feat(rust): wire CodeCompressor + Kompress into live-zone dispatch + gate (3/3)

[RubenAAA]

Description

Wires the two ported compressors (Kompress + CodeCompressor) into the live-zone dispatcher and adds the Kompress opt-in gate. This is PR 3 of 3 splitting #1143 — the only behavior-changing piece. dispatch_compressor now routes SourceCode -> CodeCompressor and PlainText -> Kompress, filling the slots it had reserved with TODOs.

Stacked on PR 1 (Kompress) and PR 2 (CodeCompressor). Until those merge, this PR's diff against main also includes their content; review #1 and #2 first. Once they land, this PR shrinks to the dispatch wiring + flag (~250 lines, mostly the routing test).

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation update
• Performance improvement
• Code refactoring (no functional changes)

Changes Made

• headroom-core: transforms/live_zone.rs — dispatch_compressor routes SourceCode -> CodeCompressor (always-on; grammars are statically linked, constructs in microseconds with no I/O) and PlainText -> Kompress (cache-only, gated). Adds set_kompress_enabled + warm_live_zone_compressors.
• headroom-core: transforms/kompress.rs — small additions to support the dispatcher's cache-only path.
• headroom-proxy: config.rs / main.rs — --enable-kompress / HEADROOM_PROXY_ENABLE_KOMPRESS flag (default off; Kompress carries a ~261 MB model so it is opt-in, mirroring the Python config.enable_kompress). Startup sets the gate and fires an off-request-path cache-only warm-up when enabled.
• tests: flipped/added live-zone dispatch routing tests.
• CHANGELOG.md: Unreleased entries for both compressors.

Testing

• Unit tests pass (pytest) — N/A for runtime (Rust change)
• Linting passes (ruff check .) — also cargo clippy -D warnings
• Type checking passes (mypy headroom) — N/A (Rust-only runtime change)
• New tests added for new functionality
• Manual testing performed

Test Output

# Rust suites
cargo test -p headroom-core   ->  918 passed, 3 ignored (14 suites)
cargo test -p headroom-proxy  ->  407 passed (34 suites)
cargo test -p headroom-core --test live_zone_dispatch  ->  routing tests green

# Parity still byte-exact after wiring
[code_aware_compressor] total=30 matched=30 skipped=0 diffed=0
[kompress             ] total=21 matched=21 skipped=0 diffed=0

cargo clippy -p headroom-core -p headroom-proxy -p headroom-parity -- -D warnings  ->  No issues found
cargo fmt --check  ->  clean
ruff check . && ruff format --check .  ->  All checks passed

Real Behavior Proof

• Environment: Linux (WSL2, kernel 6.6), Rust stable toolchain.
• Exact command / steps: cargo test -p headroom-core and cargo test -p headroom-proxy; cargo test -p headroom-core --test live_zone_dispatch; re-ran the parity harness to confirm wiring did not perturb output.
• Observed result: 918 core + 407 proxy tests pass; live-zone dispatch routing tests green; parity remains 30/30 code + 21/21 kompress; clippy and fmt clean. With the gate off (default), PlainText passes through unchanged; with --enable-kompress and a cold cache, dispatch still passes plain text through (Kompress unavailable) rather than blocking.
• Not tested: end-to-end against a live provider (dispatcher is covered by integration tests, not live traffic in this PR); alternate ONNX execution providers (default CPU EP here).

Review Readiness

• I have performed a self-review
• This PR is ready for human review

Additional Notes

• Split of feat(rust): port CodeCompressor + Kompress live-zone compressors to Rust (parity-gated) #1143 — PR 3 of 3. Lands the engines from PR 1 + PR 2 into the dispatcher.
• CCR markers are left to the dispatcher: both engines return the compressed string only; live-zone CCR uses the <<ccr:>> convention, so fixtures are recorded with enable_ccr=False for determinism.
• pytest/mypy checklist items are N/A — Rust change; equivalent gates are cargo test + cargo clippy, run above.

---

Update — non-blocking model load + rebase

Pushed fix(kompress): never block the request path on model load and rebased this branch onto the updated #1153/#1154.

• Non-blocking kompress() — the request-path accessor previously used OnceLock::get_or_init, so a request thread that hit a PlainText block while the ~261 MB model was still loading (or while a slow EP graph compile ran — the OpenVINO NPU compile takes ~13s+) blocked on that init and stalled the proxy. It's now a non-blocking OnceLock::get (returns None ⇒ pass through until ready); the one-time load happens only in warm_live_zone_compressors, off the request path. So an enabled-but-not-yet-warm Kompress degrades to passthrough instead of hanging live traffic.
• Rebase — from_cache/hf_cache_file moved to the engine PR (feat(rust): port Kompress ML prose compressor to Rust (parity-only, 1/3) #1153) where they belong; this branch's wiring commit reduces to the dispatch glue. Stack integrity intact (stack/1 ← stack/2 ← stack/3).

Validated downstream against a live Intel NPU run: proxy stays responsive through the NPU compile, then kompress_ready:true and prose blocks compress on-device. cargo fmt/clippy clean, dispatch + core tests green.

[github-actions]

PR governance

This PR follows the template and is marked ready for human review.

[RubenAAA]

Rebased onto the updated #1153/#1154 and added fix(kompress): never block the request path on model load — the old get_or_init accessor blocked request threads on the (slow, ~13s+ on NPU) model load and stalled the proxy; it's now a non-blocking get that degrades to passthrough until the off-path warm completes. Stack integrity intact. Description updated.

[JerrettDavis]

This PR is not merge-ready in current GitHub state (mergeStateStatus=UNSTABLE). Please update from current main, resolve any conflicts if present, and rerun/clear required CI before this can be approved.