Parallelize CI gates on 2-vCPU runners with GitHub fallback

[RyanLee-Dev]

The full CI gate currently runs sequentially on a 16-vCPU Blacksmith runner, while native and release jobs also select larger paid runner sizes. Split the gate into concurrent backend, tooling, Web unit/build and two isolated Playwright shard jobs, with a stable check job that requires every partition to succeed.

Use 2-vCPU Blacksmith Linux and Windows runners throughout CI and releases. Run macOS ARM64 on GitHub's standard macos-15 runner, and provide the OAC_USE_GITHUB_RUNNERS=true repository variable to switch all jobs to standard GitHub runners when needed. Partition Go cache keys, expose Web check targets in Make, and document the runner policy and gate ownership in the maintainer guide.

Remote validation passed for commit 765f99e262e8f7f3d07b621a41ec02c7d575402b:

• Full CI gate: all partitions and the aggregate check passed in 8m04s. The two browser shards passed all 48 + 38 tests.
• Native platforms, official API client acceptance and workflow syntax passed.
• Release build passed with offline artifacts enabled and publication disabled; artifacts are available on the run.

Fresh independent blind review of the complete diff found no substantive in-scope issues. Runner switching is manual; no quota polling or automatic switching is implemented.