#6677: fix(tts): always load fresh config for voice selection

## Problem The TTS tool was caching config at tool creation time, causing voice/provider changes to be ignored until gateway restart. When users changed the TTS voice via config (e.g. from `alloy` to `fable`), the change wouldn't take effect because the tool kept using the stale config object passed during initialization. ## Solution Changed `tts-tool.ts` to call `loadConfig()` on each invocation instead of using the cached `opts?.config`. This ensures runtime config changes (voice selection, provider changes via `/tts` command or config updates) are picked up immediately. ## Testing - Verified voice changes take effect immediately after config update - Confirmed OpenAI TTS uses correct voice from config <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR changes the TTS agent tool to call `loadConfig()` on each `execute` invocation instead of using the config captured at tool creation time, so runtime updates (voice/provider selection) are picked up without restarting the gateway. Change is localized to `src/agents/tools/tts-tool.ts`, and it feeds the freshly loaded config into the existing `textToSpeech` pipeline while keeping channel selection behavior the same. <h3>Confidence Score: 4/5</h3> - This PR is safe to merge with low risk; behavior change is straightforward and localized. - The change removes reliance on a captured config object and instead uses the existing `loadConfig()` API (which already has internal caching/invalidations). Primary concern is interface drift: `opts.config` is now effectively ignored, which may surprise callers/tests, but runtime TTS behavior should improve as intended. - src/agents/tools/tts-tool.ts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->