Address review feedback: de-duplicate file-scan proto conversions

[adriangb]

Which issue does this PR close?

• N/A — this is review feedback for Add DataSource/FileSource proto hooks and FileScanConfig serde apache/datafusion#23683 (which closes Proto: add DataSource/FileSource try_to_proto hook + port FileScanConfig serde apache/datafusion#23497), targeting that PR's branch.

Rationale for this change

Follow-up on my review comments on apache#23683. The theme was that
the new file_scan_config/proto.rs re-implements wire logic that already exists
elsewhere, so the copies can drift. This addresses that plus the naming and
deprecation notes.

Note on how to take this. Two of the four commits here are generic
plumbing that applies to main today, independent of the data-source hooks,
so I've also put them up as standalone precursor PRs against main:

• refactor(proto): put Partitioning / sort-expression serde on the types apache/datafusion#24003 — Partitioning / PhysicalSortExpr serde on the types (commit 4 here)
• refactor(proto): move PartitionedFile / FileGroup serde into datafusion-datasource apache/datafusion#24006 — PartitionedFile / FileGroup serde in datafusion-datasource (commit 1 here)

If those land first, rebasing apache#23683 on main collapses those two commits to
just the call sites in file_scan_config/proto.rs, and apache#23683 stays focused on
what it is actually about: the DataSource / FileSource try_to_proto
hooks, the FileScanConfig spine and the delegation through DataSourceExec.
That is the shape I'd prefer — this branch exists mainly so you can see the end
state, and you should feel free to take it as guidance rather than merge it.

What changes are included in this PR?

Four commits, each independently buildable:

1. Move PartitionedFile / FileGroup proto conversions into
   datafusion-datasource — the single copy of that wire logic now lives next
   to the types that own it, in a new datafusion_datasource::proto module
   (feature proto): FileRange, PartitionedFile and FileGroup all get
   inherent try_to_proto / try_from_proto. datafusion-proto's
   TryFromProto impls for those types become one-line shims delegating to
   them, and the private partitioned_file_* / file_group_* helpers in
   file_scan_config/proto.rs are gone — so the pub(crate) visibility
   question from @buraksenn's comment is moot too: these are public API on the
   types now, usable from Proto: add DataSink serialization hook apache/datafusion#23752.

   This uses inherent methods rather than moving TryFromProto into
   datafusion-proto-models first. Moving the trait on its own breaks the other
   45 impls in datafusion-proto (E0117) — they rely on it being crate-local
   because both of their types are foreign — so the trait move only works as
   part of relocating every impl to the crate that owns one of its types.
   That's worth doing as its own mechanical precursor PR (see the description of
   Add DataSource/FileSource proto hooks and FileScanConfig serde apache/datafusion#23683's parent epic); this PR doesn't block on it, and if
   the trait does move later, these inherent methods are what the relocated
   TryFromProto impls would call anyway.

2. Rename FileScanConfig::to_proto_conf / from_proto_conf →
   try_to_proto / try_from_proto, matching the hook naming convention.
   FileScanConfig also implements DataSource::try_to_proto, so the inherent
   encode method shadows the trait method for callers holding a concrete
   FileScanConfig — they differ in return type, so a mistaken call is a
   compile error rather than silent misbehavior, and dyn DataSource callers
   are unaffected. Documented on the method; the decode side has no overlap at
   all.

3. Deprecate parse_table_schema_from_proto in
   datafusion_proto::physical_plan::from_proto (it is now a pure delegation),
   point the in-crate callers at FileScanConfig::parse_table_schema_from_proto,
   and add an upgrade-guide entry — answering the "presumably will be deprecated
   in a future PR?" comment.

4. Put the Partitioning / PhysicalSortExpr conversion on the types
   (not from a review comment — an issue I hit while doing the above). The new
   file_scan_config/proto.rs carried its own copy of the Partitioning and
   PhysicalSortExprNode wire logic, which was already duplicated inline in
   RepartitionExec's hook and in datafusion-proto's
   serialize_partitioning / parse_protobuf_partitioning — i.e. this PR made
   it three copies. Applying the same principle as commit 1, the single copy now
   lives next to the types, taking the expression-level context
   (datafusion-physical-expr{,-common} already carry the proto feature):

   • PhysicalSortExpr::try_to_proto / try_from_proto
   • Partitioning::try_to_proto / try_from_proto

   So plan hooks can reach them, ExecutionPlanEncodeCtx /
   ExecutionPlanDecodeCtx now back the expression-level contexts and hand one
   out via expr_ctx() — useful beyond this PR, since any plan hook can now
   pass its ctx to expression-level conversions. FileScanConfig,
   RepartitionExec and datafusion-proto's central serializer all route
   through the type methods, retiring serialize_range_partitioning,
   serialize_range_split_point, parse_protobuf_range_partitioning and
   parse_protobuf_range_split_point. This one is the easiest to split into its
   own PR against main if you'd rather keep this branch focused.

The wire format is unchanged throughout. Behavior differences, both in commit 4:
out-of-range partition counts now error instead of wrapping or panicking on an
unwrap, and a missing sort-expression child reports which field is missing
instead of "Unexpected empty physical expression".

Note: a pre-existing round-trip asymmetry (not fixed here)

While adding tests for PartitionedFile::try_to_proto I found that
PartitionedFile statistics do not round-trip cleanly, on main today and
independent of this branch — filed as apache#23998. Not fixed here
since it changes decode semantics.

Are these changes tested?

Yes.

• New unit tests in datafusion_datasource::proto for the PartitionedFile /
  FileGroup round trips and the invalid-path error.
• The existing datafusion-proto tests now exercise the delegating shims (and
  would catch an accidental self-recursion in them).
• FileScanConfig serde tests from this branch cover all four partitioning
  variants through the shared ctx helpers.
• Full workspace suite: 10084 passed / 0 failed
  (cargo test --profile ci --workspace --lib --tests --features avro,json,backtrace,extended_tests,recursive_protection,parquet_encryption),
  plus cargo fmt, ci/scripts/rust_clippy.sh and
  ci/scripts/doc_prettier_check.sh clean.

Are there any user-facing changes?

Yes, all additive or deprecating:

• New public try_to_proto / try_from_proto on FileRange,
  PartitionedFile and FileGroup (feature proto on
  datafusion-datasource).
• New try_to_proto / try_from_proto on Partitioning and
  PhysicalSortExpr, and expr_ctx() on the ExecutionPlan encode/decode
  contexts.
• datafusion_proto::physical_plan::from_proto::parse_table_schema_from_proto
  is deprecated (documented in the 55.0.0 upgrade guide).
• The FileScanConfig helpers added by this branch are renamed before release,
  so nothing published changes name.