fix(provider): recover claude startup before first prompt

[astyfx]

Summary

Starting a Crane-originated task from the prepared Stave window could leave the first provider turn with no response when Claude's streaming query closed during MCP readiness before it consumed the queued prompt. This change recovers that startup-only close and preserves the terminal classification.

Changes

Fixes

• provider: retry a Claude stream that ends after initialization before initial turn work, with the initial prompt preloaded
• provider: fail fast if the initial queue closes before accepting the prompt
• lifecycle: preserve explicit aborts and prevent returned adapter terminal events from masking timeout failures

Breaking Changes

• None.

Test Evidence

• bun run test:ci
• 376 isolated test files passed, including the Claude startup recovery and provider lifecycle regressions
• TypeScript checks and desktop builds passed as part of the CI gate

Notes

• Recovery is limited to the startup-only stream state and runs at most once, so an already active provider turn is not duplicated.
• This is a provider-runtime change; no renderer, preload, or IPC payload contract changed.

[upwind-code-us]

Upwind Code Scan - ✅ Proceed with Deployment

0 newly introduced vulnerabilities · 0 resolved · 0 total in this PR vs main

View full analysis in Upwind Console

Scan completed in 19s

Scan history (1 scan)

Commit	Scanned at	New	Resolved	Net
ce3378b <	2026-08-07 13:44 UTC	0	0	0

Last scanned: ce3378b · 2026-08-07 13:44 UTC

[upwind-code-us]

Upwind IaC Scan - ✅ Proceed with Deployment

0 newly introduced misconfigurations · 0 resolved · 0 total in this PR vs main

View full analysis in Upwind Console →

Scan completed in 3s

Scan history (1 scan)

Commit	Scanned at	New	Resolved	Net
ce3378b <	2026-08-07 13:44 UTC	0	0	0

Last scanned: ce3378b · 2026-08-07 13:44 UTC

[heath-s]

Approved