test(coverage): drive server/handlers to ≥95%

[mastermanas805]

Summary

• internal/server: 54.8% → 98.9% statement coverage
• internal/handlers: 94.1% → 100% statement coverage

What changed

• Added server_coverage_test.go with dedicated-backend dispatch tests for every gRPC RPC (Provision/Deprovision/StorageBytes/Regrade) across postgres, redis, mongo, and queue.
• Introduced a PoolClaimer interface seam (one-method, *pool.Manager still satisfies it) and a test-only SetPool so the pool-hit / pool-error / pool-miss branches in provisionPostgres/Redis/Mongo/Queue are exercisable without a real *pgxpool.Pool. Nil-pointer normalisation preserves production behaviour.
• Added end-to-end gRPC round-trip via bufconn for Provision/Deprovision/StorageBytes plus the empty-token guard.
• Covered remaining branches: callBackendVoid breaker-open, New() k8s + MinIO + dedicated-DSN constructors, teamIDFromContext via metadata, PostgresBackend/Breakers accessors, mapError invalid-arg path, pool-token-marker stripping in regradeRedis.
• Handlers: added the >80-char error truncation test on platformDBCheck and a whitebox test for the defensive h.pool == nil branch.

Test plan

• make gate green locally
• Joint go test ./internal/server ./internal/handlers -coverprofile=… reports ≥95% on each
• go test ./... green across all packages
• No production behaviour change — the *pool.Manager → PoolClaimer switch normalises nil pointers to nil interfaces in NewWithBackends

🤖 Generated with Claude Code