feat(google): service-account auth for TTS + Imagen (Vertex AI), fix false-availability bugs

[shubham21155102]

What & why

Google's google_tts and google_imagen tools advertised service-account auth (GOOGLE_APPLICATION_CREDENTIALS) but only ever authenticated with an API key string, so anyone with a Google service-account JSON (common in GCP shops) could not use either tool. Along the way this surfaced two related false-availability bugs. Fixes #131.

Changes

• tools/google_credentials.py (new) — lazy google-auth helper that mints an OAuth Bearer token from the service-account JSON and resolves the GCP project id. google-auth is imported lazily so a missing dep is an actionable runtime error, not an import failure.
• google_tts — when only a service account is configured, authenticate Cloud Text-to-Speech with a Bearer token instead of ?key=. get_status() now reports available only when a path that actually works is configured (previously it over-reported on GOOGLE_APPLICATION_CREDENTIALS and then failed at execute()).
• google_imagen — service-account auth routes to Vertex AI ({location}-aiplatform.googleapis.com), since the AI Studio endpoint is API-key-only. Project resolves from GOOGLE_CLOUD_PROJECT or the key file; region from GOOGLE_CLOUD_LOCATION (default us-central1). The existing API-key → AI Studio path is unchanged.
• _load_dotenv (both base_tool.py and tool_registry.py) — strip inline comments (quote-aware). Previously GOOGLE_API_KEY= # comment parsed the comment as the value, so after cp .env.example .env every keyed tool falsely reported available.
• requirements.txt — add google-auth.
• .env.example — document GOOGLE_APPLICATION_CREDENTIALS / GOOGLE_CLOUD_PROJECT / GOOGLE_CLOUD_LOCATION.
• .gitignore — never commit GCP service-account key files.

Testing

Verified locally with a real service account (no API key set):

Tool	Auth path	Result
google_tts	service account → Cloud TTS Bearer	valid MP3 (24 kHz mono)
google_imagen	service account → Vertex AI Bearer	valid 1408×768 PNG, model imagen-4.0-generate-001

Availability honesty: blank .env → both unavailable; JSON set → available and actually generates. Invalid JSON → clean actionable error, no crash.

Full suite: 334 passed. The only 2 failures pre-exist on clean main (unrelated doubao provider set + character-animation runtime-doc contract) — this PR adds no new failures.

Notes for reviewers

• The .env parser change is global. It's quote-aware and only strips a # at the start of the value or after whitespace, matching the python-dotenv convention. Values needing a literal # should be quoted.
• Imagen on Vertex uses the same imagen-4.0-* model ids as the AI Studio path; confirmed working against a live project.