feat(stt): add Yandex SpeechKit STT provider

[ibaldr89]

What does this PR do?

Adds Yandex SpeechKit Sync API as a seventh STT provider. Yandex SpeechKit
has best-in-class Russian speech recognition accuracy (~97% WER). It natively
accepts OGG/Opus, which is the default voice format sent by Telegram and
WhatsApp — meaning no re-encoding is needed for the most common gateway use case.

The provider slots into the existing STT abstraction with zero changes to
callers. It lives entirely within tools/transcription_tools.py.

Related Issue

N/A

Type of Change

• ✨ New feature (non-breaking change that adds functionality)

Changes Made

• tools/transcription_tools.py — _transcribe_yandex() provider function;
  three helper functions (_get_audio_duration_seconds, _get_audio_channels,
  _ensure_mono_for_yandex); wired into _get_provider() explicit check and
  auto-detect fallback chain; added to transcribe_audio() dispatcher
• tests/tools/test_transcription_tools.py — 22 new unit tests covering the
  provider function, helper functions, dispatcher integration, and auto-detect
  logic; also adds the missing XAI_API_KEY delenv to the clean_env fixture
• cli-config.yaml.example — documents the stt.yandex config sub-section
  and updates the provider list comment (which was missing xAI too)

How to Test

1. Set YANDEX_API_KEY=<your key> in ~/.hermes/.env
2. Send a voice message on Telegram — the gateway will transcribe it via
   Yandex SpeechKit automatically (auto-detect picks Yandex when no local
   faster-whisper is installed)
3. Or force it explicitly: add stt: { provider: yandex } to ~/.hermes/config.yaml
4. Run unit tests: pytest tests/tools/test_transcription_tools.py -v -k "Yandex or yandex or EnsureMono or AudioChannel or AudioDuration"

Expected: all 22 new tests pass without a real API key (all HTTP calls are mocked).

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (feat(stt):)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes
• I've tested on my platform: Ubuntu 24.04

Documentation & Housekeeping

• I've updated cli-config.yaml.example with the new stt.yandex config keys
• N/A — no CONTRIBUTING.md or AGENTS.md architecture changes
• Cross-platform: all open() calls use binary mode; subprocess.run uses list form with check=False; no POSIX-only signals or file modes used
• N/A — tool schemas unchanged (STT is internal, not exposed as an LLM tool)

[teknium1]

Thanks for the thorough implementation and focused test coverage.

This automated hermes-sweeper review is closing this under the standing in-tree-provider-integration policy: third-party vendor integrations should ship as standalone plugins rather than adding vendor-specific code to the core tree.

• The PR adds the Yandex-specific handler directly to tools/transcription_tools.py:871 (ff3364c148d601f929f3095ec453ffe130b1ab8a).
• Current main already provides the intended route: TranscriptionProvider explicitly supports new STT backends without modifying tools/transcription_tools.py (agent/transcription_provider.py:22-35), and transcribe_audio() dispatches non-built-in providers through that plugin surface (tools/transcription_tools.py:1713-1737).
• Please publish this as a standalone Yandex SpeechKit STT plugin using ctx.register_transcription_provider(); users can install it under ~/.hermes/plugins/ or through a pip entry point.

---

Closed as not-planned per standing maintainer policy (in-tree-provider-integration). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.