Fix Vertex OpenAI-compatible routing and countTokens

[LeenHawk]

Summary

• route Vertex OpenAI Chat Completions passthrough requests to the official endpoints/openapi/chat/completions endpoint
• flatten Vertex countTokens generateContentRequest bodies before sending to Vertex publisher models:countTokens
• update the gproxy-protocol submodule pointer to the latest main
• add a root pnpm manifest so Dependabot can evaluate the root pnpm-lock.yaml

Fixes #92

Tests

• cargo test -p gproxy-channel --test routing_alignment
• cargo clippy --workspace --all-targets
• pnpm install --frozen-lockfile