fix(core): retry provider failures for about 84s with 10s gap cap - #49441

Summary

• raise the session retry allowance from 4 to 10 retries (11 attempts)
• cap each backoff gap at 10s via Schedule.min with spaced("10 seconds"); the exponential base stays 2s and jitter stays ±20%
• rewrite the runner tests that encoded the old count to derive their expectations from one RETRY_GAPS constant

Schedule

No retry-after header, every attempt failing:

Total wait 84s nominal (67–101s with jitter), previously 30s with a 16s longest gap. retry-after handling is unchanged: the gap is max(computed, retryAfter) with the existing 15 minute ceiling, so a provider can still stretch a single gap past 10s when it asks to.

Verification

• packages/core: session-runner 210 pass; full suite passes except the pre-existing local shell/env failures that also fail on v2
• bun run check