fix(conc): scope channel hangup/answer callbacks to fix per-call leak (#121)

[ryanmurf]

Fixes #121.

Problem

HANGUP_CALLBACKS / ANSWER_CALLBACKS in asterisk-core/src/channel/mod.rs
were append-only Vecs. Every inbound INVITE registered two per-call
closures (event_handler.rs:768/:775) keyed to one channel's unique_id
and never removed them; ConfBridge joins (confbridge.rs:1465) and Local
pairs (local.rs:301) added more.

• Unbounded memory: ~2 leaked closures per call. Over the M5 500-call soak
  (and beyond) this grows without bound → OOM. The M4 exact-baseline harness
  never caught it — it counts the transaction/channel/call-state/NOTIFY maps,
  not these registries.
• O(n²) fire cost: fire_*_callbacks scans the whole vector on every
  hangup/answer; the Nth event scans ~2N stale closures.

Fix

• Key both registries by a monotonic id: HashMap<u64, Callback>.
• Add a scoped RAII API — register_{hangup,answer}_callback_scoped returning a
  #[must_use] handle whose Drop unregisters the entry.
• Migrate the four per-call sites to the scoped API, holding the handle for the
  call/leg lifetime (moved into the per-call task so it drops at task end).
• Keep the permanent register_* API for process-lifetime subscribers (CDR).
• Add registered_{hangup,answer}_callbacks() count accessors so the soak
  harness can assert the registries return to baseline.

Locking is unchanged (fire_* still invokes closures under the registry lock);
documented the pre-existing invariant that a closure must not register/drop a
handle synchronously from inside a callback.

Tests (red-capable, asterisk-core)

• scoped_{hangup,answer}_callback_unregisters_on_drop — baseline+1 while held,
  back to baseline on drop.
• scoped_hangup_callback_fires_while_held_and_never_after_drop — functional:
  fires once while held, never after drop (even for a reused unique_id).
• soak_scoped_callbacks_return_to_baseline — 500 register/drop cycles return
  both registries to baseline (mirrors the M4 exact-baseline discipline).

RED control

Neutralizing both Drop impls (no-op removal — the pre-fix leak) fails all four:

scoped_hangup_callback_unregisters_on_drop:   left: 2  right: 1   "return the registry to baseline (leak)"
scoped_hangup_callback_fires_..._after_drop:  left: 2  right: 1   "dropped callback must not fire again"
soak_scoped_callbacks_return_to_baseline:     left: 502 right: 2  "hangup registry leaked across 500 simulated calls"

Gates

• cargo test --workspace --exclude pjsip-shim — 4362 passed, 1 ignored.
• cargo clippy --workspace --exclude pjsip-shim -- -D warnings — clean.
• Rust 1.97.0. No cargo fmt run.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XH6jhrG4ZCL8xfCktK4Zms