[max-charlamb]

Summary

Fixes dotnet/runtime#136190.

The SOS test treated dumpgcdata's pre and post pin statistic as a pinned-object count. It actually records a GC pinned-plug bookkeeping condition. The expected value depended on a runtime-owned pinned sentinel, not the debuggee's pinned 1 MB array. Moving that sentinel to the frozen heap in dotnet/runtime#135326 exposed the faulty assumption; changing the expectation to zero would preserve the same problem.

Replace this incidental oracle with a controlled GC operation:

• Explicitly request a forced, blocking, compacting full collection between two stop points.
• Check that each heap's induced compacting GC counter increases from its observed baseline, without assuming an initial or final count.
• Preserve the per-heap GC statistics coverage for [DAC] fix dac_gc_heap marshalling runtime#119393, including server GC, and retain the .NET 8/9 exclusion for that DAC marshalling defect.
• Update the debuggee, stop-point catalog, test names, and coverage documentation together.

No runtime or cDAC implementation changes are needed.

Validation

• Targeted SOS.Tests and debugger-host builds passed.
• All 15 revised cases passed on Windows x64 across .NET 10/11, CDB and dotnet-dump, live/dump targets, workstation/server GC, and applicable legacy/cDAC configurations.
• Negative control: temporarily removing the explicit collection caused the .NET 11 server-GC cDAC case to fail because the counter remained zero. Restoring the collection made the matrix pass again.

[max-charlamb]

/ba-g CI previously passed and blocks clean CI in runtime