End-to-end validation of SpdSayBackend on Linux

[digitalby]

SpdSayBackend is implemented and unit-tested for rate mapping, but it has never been exercised end-to-end against a real speech-dispatcher daemon with acc claude.

Validate:

• spd-say picks up utterances correctly
• Barge-in (Ctrl-\) actually interrupts speech mid-sentence
• Rate mapping (WPM → spd-say -100..+100) sounds right to a blind user
• Works under orca-style setups where speech-dispatcher already has a driver loaded