docs: make README a concise user guide

[hartsock]

Summary

Reorganize README.md into a concise, user-facing landing page and move the
detailed freeze rationale + history into a new docs/STABILITY.md. Documentation
only — no runtime, public API, wire format, feature-flag, dependency, or
release-config changes.

• The README is reorganized around installation, quick starts, safety boundaries,
  and feature status — a new user reaches a working example before internal design
  history.
• The merged store feature is now documented against the real API
  (AddressedBytes, VerifiedStore, get_node, put_node, put_checked,
  decode_verified_bytes) — not the old PR description.
• VerifiedStore is presented as the recommended verified facade.
• Detailed stability decisions moved to docs/STABILITY.md (with issue provenance).
• Removed stale test counts, coverage claims, and obsolete API names (get_typed,
  insert(id, bytes)).

Before → after (README structure)

Before	After
Title + long IPLD essay + explicit CID-build prose	Title + one-paragraph pitch + fixed-profile table
"IPLD-native" deep dive up front	folded into the intro table
"Presentation contract (FROZEN)" + hex debate inline	compact 4-form table; hex rationale → docs/STABILITY.md
Usage (single trait example)	Rust quick start (trait + content_id + verify)
Python block	Python quick start (mirrors tests/test_readme.py)
—	Choosing a construction path table (safe vs unchecked)
—	Experimental features: merkle + accurate store (VerifiedStore example, trust boundary)
"Alpha status — bytes are NOT frozen" + 10-item SETTLED ledger	Status & stability table + one paragraph → docs/STABILITY.md
CI / release-pipeline mechanics inline	short Development + link to RELEASING.md

docs/STABILITY.md (new): the frozen 0.1.x contracts rewritten as contract
statements by area (CID profile, presentation, serde repr, checked/unchecked
construction, ContentError policy, verify/ensure_content_id, crate-root
exports, MSRV/edition, no-rehash digest bridge) + the experimental merkle/store
exclusions, with #3–#10/#71 provenance.

Trust-boundary accuracy (store)

The README's store section deliberately does not claim: every backend is
grow-only; a successful backend write was durably persisted; raw get proves
canonicality; inherent method resolution is intrinsically unskippable; the formal
properties are mechanized; or the collision branch has an executable Rust test. It
states the seam theorems vs backend refinement laws split and links the deferred
formal targets (#71) — matching the module docs in src/store.rs.

Validation

• cargo test --doc --all-features — green.
• The three README Rust examples (quick start, merkle, store) compiled and
  run against main in a temporary tests/ harness (removed, not committed).
• The README Python block is kept verbatim-mirrored by tests/test_readme.py
  (unchanged), which CI's python job (maturin develop + pytest) runs.
  Limitation: the Python extension is not built in the authoring environment, so
  every imported symbol was cross-checked against content-addressable-py/src/lib.rs
  (class exported as ContentId; to_bytes/from_bytes/parse/
  from_canonical_bytes/from_blake3_content_digest/digest_bytes/digest_hex
  present; content_id/to_canonical_dagcbor/from_canonical_dagcbor module funcs)
  rather than executed locally.
• Adversarial scan for stale phrases (get_typed, insert(id, bytes), coverage/test
  counts, cannot opt out, unskippable, global "bytes are frozen/unfrozen") — clean.
• Relative links (docs/STABILITY.md, RELEASING.md) resolve.

Out of scope

No code, API, wire format, feature flags, tests, dependencies, or release
configuration changed. docs/MERKLE_CATALOG.md / docs/MERKLE_ROADMAP.md untouched.