Use PHP SDK 2.0.10 in the standard-workflow soak

[rmcdaniel]

Change

Update the PHP capacity/soak consumer from published SDK 2.0.1 to 2.0.10, including its lockfile, adapter identity, suite artifact coordinates and exact-version readiness checks. Other dependencies, SDK languages, Server limits, deadlines, workloads and measurement semantics are unchanged. Existing results retain their original suite hash/source/artifact tuple.

Verification

• Published-package smoke passed: https://github.com/durable-workflow/sdk-php/actions/runs/34616341682 (Server 2.1.0).
• Installed 2.0.10 backpressure reproduction passed: 60 simulated seconds, 60 workflow polls, 59 activity polls, 59 query polls, 11 heartbeats, graceful shutdown. Older installed negative controls had no activity/query progress.
• Composer reports no dependency advisories; suite validation passes all nine cells.
• Local combined Python contract tests were toolchain-blocked for PHP/Cargo; public CI runs the complete toolchain and real standard-workflow soak smoke.
• Initial CI correctly caught stale suite metadata; the latest commit aligns those pins rather than bypassing validation.

Consumes durable-workflow/sdk-php#66; related #137. The daily paid soak remains paused. This is not proof that the SDK caused all six prior timeouts or two-hour endurance/capacity qualification. No Server runtime image release is needed for this benchmark-only change.

[rmcdaniel]

Reviewed the final five-file diff: only the PHP SDK lock and matching capacity artifact/readiness coordinates change to published 2.0.10; other dependencies, workload, admission limits and deadlines are unchanged.

All checks passed on head 03b766f41fd452980e751c9af6b0156fec8d576e. Server Perf run exercised the pull-request merge commit with Workflow 2.0.12 and the source-built Server: 11/11 DW Standard Workflow v1 executions completed with installed PHP SDK 2.0.10, zero errors; 1,000 synthetic starts accepted; 25 healthy resource samples; final tracked polling/server cache keys zero. The feature suite and complete contract/toolchain checks also passed.

This is a 120-second source regression smoke, not published-Server capacity or two-hour endurance qualification. The summary still has 22 ready synthetic tasks, so it is not evidence of a completely empty task backlog. The earlier long-soak timeouts remain under investigation in #137, and paid scheduled runs remain disabled.