Implement remaining dictation mode combinations

[acgxv]

Context: The STT pipeline spec notes only local stream + local second pass is implemented and all other mode combinations are TODO.

Scope:

• Define and implement the remaining supported mode combinations.
• Ensure mode routing is correct for each combination.
• Keep the UI contract stable (live_render, segment_end).

Reference: docs/spec/core/stt_dictation_pipeline.md (Implementation Status).

Acceptance:

• Document which mode combinations are supported.
• Implement routing for the missing combinations.
• Add any required tests or validation notes in the spec.