[ZhangHanDong]

Fragmented Studio responses could extend a nominal five-second timeout, while temporary WebSocket waits produced generic I/O errors. This change keeps one absolute deadline through each TCP read/write, resumes the existing upgrade state, and identifies upgrade, Hello, command, and capture failures. Input and capture requests are never replayed after uncertainty.

Integration status: the complete change is included in #336, which targets main directly. This Draft remains available to review the narrower diff; it does not need to merge before #336. Validation below applies to this branch's revision. Final integration validation and current limitations are recorded in #336.

Validation:

• 42 harness tests passed, including fragmented responses, slow writes, query/build correlation, no input replay, and retained failure evidence.
• Scoped build and Clippy passed; existing warnings remain. All 11 task-contract scenarios and the file boundary passed; a deliberate deadline-property mutation failed and the restored implementation passed.
• Independent socket probes confirmed the upgrade deadline at 5.007 seconds, a separate Hello deadline, immediate HTTP 403 rejection, and one connection across fragmented upgrade/Hello.
• In the combined local Makepad 5ab3161 runtime, the built harness captured the selected app and switched to Home successfully. The combined workspace passed 793 tests with 5 ignored, release build, and Clippy; those combined results also include approval changes outside this PR.

The intermittent live Studio stall's cause remains unproven, and the separate Computer Use window-discovery error remains unresolved. This is not full approval/provider E2E acceptance. CI availability depends on the stacked base and draft workflow filters; skipped checks are not passing test evidence.