fix: extend TTFT histogram upper-tail buckets

[pjb157]

Summary\n- preserve the existing time-to-first-token histogram boundaries through 10 seconds\n- add a sparse upper tail through 120 seconds\n- verify the exported boundary set with a focused regression test\n\n## Cardinality\nThis adds eight classic histogram bucket series per populated label set. The spacing widens as latency increases to improve upper-tail resolution without introducing a dense upper range.\n\n## Rollout\nBecause the classic histogram keeps its existing metric name, lookback queries that span deployment temporarily mix the old and new bucket schemas. The 24h, 7d, and 30d views may show distorted quantiles until each window contains only post-deploy samples.\n\n## Verification\n- cargo fmt --all --check\n- cargo test -p dwctl metrics::gen_ai::tests\n- SQLX_OFFLINE=true cargo clippy -p dwctl --lib -- -D warnings

[cloudflare-workers-and-pages]

Deploying control-layer with    Cloudflare Pages

Latest commit:	9440c06
Status:	✅  Deploy successful!
Preview URL:	https://bb5f563d.control-layer.pages.dev
Branch Preview URL:	https://peter-extend-ttft-histogram.control-layer.pages.dev

View logs

[Copilot]

Pull request overview

Extends the Prometheus classic histogram bucket boundaries for GenAI “time to first token” (TTFT) to better capture slow streaming responses while keeping the existing OTel-derived buckets through 10 seconds, and adds a focused regression test to lock in the exported boundary set.

Changes:

• Extend gen_ai_server_time_to_first_token_seconds histogram buckets with a sparse upper-tail out to 120 seconds (preserving existing buckets through 10s).
• Add a unit test that asserts the gathered TTFT histogram exports the expected upper-bound list.