fix(soniox): widen final span to all timed tokens, never regress end_time

[rkfshakti]

Fixes #6885

Problem

A FINAL_TRANSCRIPT emitted at a natural endpoint can carry a start_time/end_time span far shorter than the speech its own text represents — e.g. "He's trying." with a 0.18s span.

Root cause

_TokenAccumulator.update() took start_time from the first token that carried start_ms and overwrote end_time from every token that carried end_ms:

if "start_ms" in token and not self._has_start_time:
    self._has_start_time = True
    self.start_time = float(token["start_ms"])
if "end_ms" in token:
    self.end_time = float(token["end_ms"])

Two failure mechanisms (both confirmed against the fake-WS harness):

1. Late timing: leading tokens arrive without start_ms/end_ms, only a trailing token has them → the span collapses onto that trailing token.
2. Regressing end: a trailing token whose end_ms is smaller than an earlier one pulls the end backwards.

merged_speech_data() (the interim path) already guards both with min(...)/max(...); the final path did not.

Fix

• start_time = earliest start_ms seen (not just the first)
• end_time = latest end_ms seen (never regresses)

Regression tests cover both mechanisms. Verified: 2 of 3 new tests fail on the old code, all 31 soniox tests pass with the fix.

[CLAassistant]

All committers have signed the CLA.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.