[Extensibility 4/4] feat: Vertex AI provider (stacked on #259)

From the PR description

Stacked on #259 - merge that first; until it merges, this diff shows its commit too. This PR's own change is the single Vertex commit.

Summary

Google models via Vertex AI without touching call sites: setupVertexAI() re-registers the "gemini" provider id so the same Gemini model IDs route through Vertex AI - enterprise billing, IAM-gated access, Workload Identity instead of a distributed API key. Zero behavior change unless configured: nothing calls setupVertexAI() by default.

Changes

• providers/vertexAI.ts (new): streams/completes via @google/genai with vertexai: true (ADC, VERTEX_AI_PROJECT / VERTEX_AI_LOCATION); tool calls, thinking deltas, abort supported. No new dependencies - @google/genai is already in backend/package.json.
• __tests__/vertexAI.test.ts: 7 tests (registration, model matching, re-registration).

Why

AI Studio API keys don't fit enterprise deployments needing project-scoped billing, VPC-SC/data-residency compliance, or keyless Workload Identity.

Testing

On this branch: backend npm test - 286 passed, 5 skipped (includes the 7 new tests). tsc clean.

Provenance

Both files byte-identical to the code running in amal66/mike (main). Details in amal66#48.

🤖 Generated with Claude Code