fix(runtime): route parse error slots through a silent context probe

[slepp]

Why

Parse error-slot routing (set_error/clear_error/get_error in hew-runtime/src/parse_error_slot.rs) identified the current actor via the diagnostic extern hew_actor_current_id(). When no execution context is installed, that extern writes the EXECUTION_CONTEXT_NOT_INSTALLED sentinel into LAST_ERROR as a side effect — so merely routing an error slot could corrupt the very error state the slot exists to protect. A prior parse error (or a clean null slot) got clobbered with an unrelated context diagnostic on non-actor threads.

Fixes #2658

What

• runtime: added hew_actor_current_id_silent(), a pub(crate) probe built on the non-clobbering execution_context::current_context() and the shared actor_id_from_context helper — same actor-id semantics, no LAST_ERROR write. All three parse error-slot routing sites now use it. The diagnostic extern hew_actor_current_id() is untouched: same #[no_mangle] extern "C" signature, same sentinel-write contract (callers such as the execution-context tests explicitly consume that write).
• tests: a divergence test pins both contracts — silent probe returns -1 and leaves hew_last_error() null; diagnostic extern returns -1 and writes exactly the EXECUTION_CONTEXT_NOT_INSTALLED constant. Four non-actor regressions cover the error slot preserving a prior error and staying null from clean. A WASM scheduler parity test drives a real mailbox → enqueue → hew_sched_run dispatch and asserts both probes agree inside dispatch and stay silent outside it.

Test

• silent_probe_diverges_from_diagnostic_on_missing_context (actor.rs) — mutation-verified: reverting the silent probe to require_current_context() fails this test plus all four non-actor slot tests
• nonactor_* regression tests in parse_error_slot.rs (prior-error preserved; slot stays null from clean)
• self_api_sees_current_actor_during_dispatch extended in scheduler_wasm.rs for silent/diagnostic probe parity on the WASM scheduler path

Quality Checklist

• PR title, body, and commit messages avoid internal-only orchestration/model vocabulary
• No new .ok()? or unwrap_or_default() in codegen without // JUSTIFIED comment
• New allocations have cleanup paths for sync, async, and actor shutdown contexts
• Serialization changes include round-trip encode/decode tests (n/a — no serialization change)
• New runtime features have WASM implementation or // WASM-TODO(#NNN): marker, and new hew_* exports are classified in scripts/jit-symbol-classification.toml (silent probe is pub(crate), not an export; WASM path covered by the parity test)
• No duplicated logic — the silent probe reuses current_context() and the shared actor_id_from_context helper