perf(webhook): prune request caches incrementally

[teknium1]

Summary

Webhook POST handling now prunes request caches incrementally instead of rebuilding TTL lists/dicts on every accepted request.

Changes

• gateway/platforms/webhook.py: store per-route rate-limit windows as deques and prune expired timestamps from the left.
• gateway/platforms/webhook.py: check idempotency TTL for the current delivery ID directly, with occasional bounded stale-cache cleanup.
• gateway/platforms/webhook.py: track delivery-info creation order so TTL pruning only walks stale entries until the first fresh one.
• tests/gateway/test_webhook_adapter.py: cover incremental rate-limit pruning, per-delivery idempotency expiry, and ordered delivery-info pruning.

Validation

Check	Result
Focused regression set	5 passed in 0.17s
Webhook adapter suite	68 passed in 0.50s
Targeted webhook files	82 passed in 0.7s
Syntax/diff checks	py_compile + git diff --check passed

Notes

• Preserves HMAC auth, rate limiting, idempotency, deliver_only, and delivery-info retention semantics.
• The untracked .plans/speed-optimization-audit.md remains intentionally excluded from this PR.

Infographic

[github-actions]

🔎 Lint report: hermes/webhook-ttl-pruning vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 10893 on HEAD, 10893 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5725 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.