refactor(sdk): convert bind-macro error/level shims to From impls

[lgahdl]

Summary

Fixes #264, #286. Follow-up to #185's chain-log WIT boundary cleanup, which converted assemble_log/project_chain_log into From impls under the "verb-named conversions should be From" idiom.

nexum-sdk's bind_host_via_wit_bindgen! still emitted four pure, infallible type-to-type conversions as named functions (convert_chain_err, convert_fault, sdk_fault_into_wit, convert_level — the issue's text names three of these under an older pre-rename naming, convert_err/sdk_err_into_wit; the fourth, convert_chain_err, was added since but is the same shape), so call sites read .map_err(sdk_fault_into_wit) instead of plain ?.

All four become From impls instead. This is orphan-rule-allowed because the macro always expands inside the per-cdylib module crate, where the wit-bindgen-generated type is local and either sits as Self or covers the From<T> type parameter — exactly the mechanism #185 already established.

Also closes #286: that issue asked for exactly From/TryFrom between the wit-bindgen Fault and the SDK Fault — filed against the pre-shim state where call sites converted "by hand" (before convert_fault/sdk_fault_into_wit existed as named functions). The impl From<nexum::host::types::Fault> for $crate::host::Fault and its reverse added here are exactly that conversion, done idiomatically. No separate work needed.

Ripples to:

• Every module lib.rs call site: twap-monitor, ethflow-watcher, http-probe, stop-loss, price-alert, balance-tracker — .map_err(sdk_fault_into_wit)? becomes plain ? (the enclosing fn's Err type already is the wit-local Fault, so ? auto-converts via the new From impl).
• shepherd-sdk's own bind_cow_host_via_wit_bindgen!, whose convert_cow_err called convert_fault(f) internally — updated to f.into(). (convert_cow_err itself is left as a named function; it's a different file/pattern not named by this issue.)
• Doc comments across all of the above that named the old functions.

No behavior change — pure ergonomics/idiom, as scoped in the issue.

Test plan

• cargo build --release --target wasm32-wasip2 for all 12 guest modules — clean
• cargo clippy --workspace --all-targets --all-features --locked -- -D warnings — clean
• cargo fmt --all -- --check — clean
• cargo test --workspace --all-features --locked — full suite green, all doctests pass
• Independent second-pass review (orphan-rule reasoning, dangling call-site check, doc accuracy) — no findings

[mfw78]

Closing as superseded by #452, which lands the same #264 refactor - the From impls for the bind-macro fault/level shims, in the same two files (crates/nexum-sdk/src/wit_bindgen_macro.rs, crates/shepherd-sdk/src/wit_bindgen_macro.rs) plus the same module call sites - on the M3 Videre train.

#358 predates the train and targets develop directly; per review (#452), keeping one avoids the duplicate work and a merge conflict once either lands. #264 will close when #452 lands.

Reopen if landing this directly on develop (ahead of the train) is preferred.