feat: record time to first token on the OpenTelemetry chat span

[cpsievert]

Ports tidyverse/ellmer#1142.

Summary

Streamed responses now record the time to first token (in seconds) as the gen_ai.response.time_to_first_chunk attribute on the OpenTelemetry chat <model> span.

• chatlas/_otel.py: record_chat_ttft_attr() sets the attribute on recording spans, given a start timestamp.
• chatlas/_chat.py: both sync and async streaming loops capture stream_start just before chat_perform() (which blocks until headers arrive, so TTFT is measured from request issuance) and record the attribute once, at the first chunk carrying non-empty text or thinking content (via the existing display_text() helper).
• Tests: offline mocked test asserting TTFT is recorded at the first non-empty text token (not merely the first chunk), plus TTFT assertions in the VCR streaming-span test and the async stream-iteration test.

Two small adaptations from the R implementation: time.monotonic() is used instead of wall-clock time since this measures an elapsed interval, and the mocked empty first delta uses ContentText.model_construct(text="") (as real providers do) because ContentText.__init__ normalizes empty text to "[empty string]".

Verification

pytest tests/test_otel.py — all 24 tests pass, including the new test_ttft_recorded_at_first_non_empty_text_token and the new TTFT assertions on sync/async streamed chat spans.