Enhance SQL parsing, agent tests, bench, and CI

[jamals86]

Improve parsing and robustness across the codebase:

• backend: Robust EXECUTE AS parsing: add parse_single_quoted and find_matching_close_paren helpers, use them in parse_execute_as to correctly handle quoted usernames and parentheses inside string literals, and add unit tests for escaped quotes and parenthesis-in-strings.
• benchv2: Simplify subscribe_change_latency_bench by awaiting subscription events directly (removes atomic counter and background listener), improve error handling and graceful subscription close.
• examples/chat-with-ai: switch agent to use client.executeAsUser, allow consumer group from KALAMDB_GROUP env, and add extensive integration tests and helpers (auth client, topic readiness, multi-tab streaming and failover tests) to validate streaming and agent failover behavior.
• CI (.github/workflows/release.yml): emit a docker version+commit tag using a short SHA and create a multi-arch manifest for that tag in addition to the version and latest manifests.
• docs: small Dart README link/update.

These changes tighten parsing correctness, improve test coverage for streaming/agent scenarios, make the bench more reliable, and tag Docker images by commit for traceability.