[FEATURE] Context-aware spelling for ambiguous words in voice dictation

Summary

When dictating, the transcriber sometimes picks the wrong spelling for
homophones / near-homophones. It would help a lot if, when a word is
ambiguous, the transcriber used the surrounding conversation context to
choose the spelling instead of guessing the word in isolation.

Example

I talk about "Claude Code" constantly, but the transcriber frequently
writes it as "cloud." Given how often "Claude Code" appears in the same
thread (and across my other threads), "cloud" is almost never what I
mean — the context makes the correct choice obvious.

Proposed behavior

When the speech-to-text model is uncertain between similar-sounding
spellings, weight the decision using:

• terms already used earlier in the current conversation, and
• frequently-used terms across my recent threads (a learned personal

vocabulary).

Why it matters

Repeated mis-transcription of common, high-frequency terms forces a
manual correction on nearly every message and breaks the flow of
dictation.