[Majed-allababedi]

Summary

Avoid erasing and reallocating the first std::map node when RangeSet::Add grows an existing interval without changing its start. Preserve the contained-range fast path, extend the retained node endpoint, and erase only absorbed successors. New-start/left-growing/disjoint insertion, End() validation, synchronization, and renderer/coherence ordering remain unchanged.

Adds standalone RangeSet semantics/allocation tests and registers them with CTest, kyty_tests, and the Windows U59 CI build/test lists.

This PR also updates two stale Windows INSERTQ test expectations to match the project's selected zero-upper-lane emulation behavior. It does not change production INSERTQ code. The ISA leaves these bits undefined; zero is the project-selected behavior, not a universally mandated result. The isolated test correction addresses the failures documented on superseded PR #36.

Verification

• Exact upstream main base: 5c3b39f76de85a0dc3292176ee04e85825ab47fb.
• Tested int19.4 source: c9b54080d42ca0eaddbbe5bab6714bbb53e67c1f; local Windows Release tests passed 9/9, including RangeSet, all five VM variants, memory_tracker, page_manager, and amd_cpu_emulation.
• Exact-base allocation tests distinguish old and new behavior: eligible merges allocate zero nodes while absorbed successors are still freed. The standalone suite also checks 60,000 randomized mutations and 240,000 queries against an independent byte model.
• The release source and existing independent reviews cover the tested code; this PR does not claim that remote CI has passed. Component benchmarks are not game-FPS measurements.

Release source/tag: u59-int19.4-rangeset-windows-local-v1.

[Jetsku]

Has been merged and added to a release