feat(tts): add Airy streaming provider

[fbdp1202]

Summary

• Add support for airy-tts-v1, the TTS model offered through Out of Set's Airy service.
• Stream audio over HTTP from https://api.airy.so/v1/audio/speech/stream using 24 kHz mono signed 16-bit little-endian PCM.
• Use shared HTTP client and TTS finalization helpers for WAV output, time-to-first-audio measurement, and connection diagnostics.
• Register the provider and document the optional AIRY_API_KEY environment variable.
• Validate response audio metadata and reject failed, interrupted, or incompatible streams.

Closes #609

Files changed

File	Purpose
runner/src/coval_bench/providers/tts/airy.py	Airy streaming TTS implementation
runner/src/coval_bench/providers/tts/__init__.py	Provider registry and exports
runner/src/coval_bench/registries/provider_keys.py	API key environment mapping
runner/src/coval_bench/config.py	Optional Airy API key setting
.env.example	Document AIRY_API_KEY
runner/tests/providers/tts/test_airy.py	Request, audio output, timing, authentication, and error-handling tests

Suggested benchmark voices

Please use the following voices when configuring Airy benchmark runs:

Voice	Voice ID
Female	a597bb7a98fc9ec1
Male	bdb7de5e2cdd3324

Test plan

• Airy provider unit tests: 17 passed.
• Ruff lint and format checks passed.
• mypy --strict src tests passed.
• Full pytest suite: 2,047 passed, 6 skipped due to unavailable optional Google SDKs and separate test fixtures.
• Live API smoke test: HTTP 200 over HTTP/2 with valid 24 kHz mono 16-bit WAV output.

[coval-cale]

Sup! Stoked to get you onboard! I can't figure out how to get an API key for local testing. shoot me an email at cale(at)coval(dot)dev and I'll set y'all up with a slack channel so we can chat easier.