Add an app-wide Context Profile for multimodal transcription

[SeoFood]

Context

#710 showed that multimodal transcription quality improves when the engine receives user-specific context: domain vocabulary, formatting preferences, and technical dictation rules. The first practical step has shipped in the existing Gemini plugin via #720 and #721, and the published Gemini plugin release is plugin-gemini-v1.0.18.

This issue tracks the remaining product work: make that context a first-class app-wide profile instead of burying it in one plugin implementation.

Goal

Add an editable Context Profile that users configure once and that compatible engines can consume when they support rich prompt/context input.

The profile should cover:

• About me / work domain
• Personal vocabulary and domain terms
• Dictation and formatting preferences, such as preserving CamelCase, identifiers, version numbers, CLI flags, and numeric values

Expected behavior

• Users can manage the profile from the app, not per plugin.
• Existing Dictionary terms remain usable and should feed into compatible transcription prompts.
• Engines that cannot use instruction-following multimodal context should not pretend to support the profile.
• Gemini transcription should be one consumer of this profile once the shared app-side plumbing exists.

Out of scope for this issue

• Reworking the already-shipped Gemini STT release.
• Reintroducing the closed reference PR [Reference / please don't merge] Gemini STT plugin — multimodal audio + context transcription #709 as a merge source.
• Local Gemma audio support, unless a separate implementation issue explicitly scopes it.

References

• Original idea and benchmark: Idea: treat transcription as "audio + personal context" with a multimodal LLM — it dramatically outperformed every ASR engine in my testing #710
• Gemini STT implementation: Add Gemini speech transcription #720
• Gemini plugin host-version fix: Require TypeWhisper 1.5 for Gemini plugin #721