Improve voice activation, add va status skill and improve hud rendering

[SawPsyder]

Also references #402

Summary

Improve voice activation, add va status skil and improve hud rendering

Changes

• Fix triggered voice activation state change during audio player playback (saves future state instead of changing anything directly)
• reuse background opacity for hud outline rendering
• add new skill that shows the voice activation status on the hud (this is also a community request)

Testing

• tested locally
• enable voice activation and change state during playback or generally to investigate shown state in wingman ai and the new skill

Checklist

• This PR is linked to a GitHub issue
• I have tested my changes locally
• I have rebased my branch onto develop

[Shackless]

Thanks for the PR! The core idea of the toggle fix is sound — I verified that all real end-of-playback paths reliably fire on_playback_finished, and streamed responses produce exactly one started/finished cycle, so the saved intent is applied correctly in the normal flow. The outline-alpha change also checks out (content-layer borders staying opaque is consistent with the design).

I left inline comments on the issues found during review, roughly in order of severity:

1. Race between the new playback branch and on_playback_started that can silently swallow a mute press (wingman_core.py)
2. The skill's permanent /ws connection breaks Core's offline-message queueing for the real client (mic_status/main.py)
3. The intent fix only covers the hotkey path — the /voice-activation/mute endpoint still has the old bug (wingman_core.py)
4. Icon paths containing ) never render, and the code comment about the parser is wrong (mic_status/main.py)
5. After a HUD server restart the icon comes back as a default-styled window, and failed draws are cached as successful (mic_status/main.py)
6. The connected guard in update_config drops config changes after any transient HUD hiccup (mic_status/main.py)
7. Reading Core internals via sys.modules["__main__"] bypasses the skill facade — uniquely among bundled skills (mic_status/main.py)
8. icon_size code default (96) drifted from default_config.yaml (72) (mic_status/main.py)

Minor nit not worth its own thread: (55, 62, 74) is now hardcoded in five places across hud_server; a DEFAULT_BORDER_COLOR in hud_server/constants.py would fit the existing pattern there.

[SawPsyder]

All issues resolved in newest commit.
But still converted the PR to draft to properly test the changes in my next session.

I moved away from "lets not touch the core where I dont have to" to "lets modify core to give me what I need for a clean state".
But therefore I will give it more time to test now.

[SawPsyder]

I think its good to go now. Also included some elevenlabs adjustments for playback state detection fixes

[Shackless]

Second review round — verified all 8 findings from the previous round against the new commits: all properly resolved ✓ (intent model, endpoint path, WS removal, paren regex, HUD restart/create-or-update, .connected guard, facade surface, yaml default sync). The sound_config parameter also fixes a genuine latent NameError in develop's direct-stream path — nice catch.

Four new findings below, posted inline: 1 blocker (VA doesn't resume after a naturally ending ElevenLabs playback — deterministic, both paths), 1 comment/code mismatch, 2 minor notes.

Note on the diff: the branch carries cherry-picked copies of four develop commits (condenser, voice service, config context, template renames) with different SHAs, which inflates the PR diff to 27 files. The effective diff vs. develop is 14 files. A rebase onto current develop before merge would clean that up.

[SawPsyder]

@Shackless Open for review again.