fix: resolve P0/P1 code quality issues

P0 fixes:
- ModelError.Is(): use exact matching instead of substring contains()
- shouldClearStickySession: add context param for cancellation/tracing

P1 fixes:
- TODO stubs: return 501 Not Implemented errors
- validateInstanceSignature: deduplicate to shared validateCodeSignature()
- Error messages: standardize to English only
- http.go: remove pseudo if-else with duplicate branches

backend/migrations/077_add_usage_log_requested_model.sql

@@ -0,0 +1,3 @@
+-- Add requested_model field to usage_logs for normalized request/upstream model tracking.
+-- NULL means historical rows written before requested_model dual-write was introduced.
+ALTER TABLE usage_logs ADD COLUMN IF NOT EXISTS requested_model VARCHAR(100);