Replace remaining generated platform copies with layer-specific functional workloads

[LunaStev]

The generator still builds each numbered target case from the same shared test104–test113 body plus a repeated target checksum/aggregate checkpoint. Selected cases were strengthened in #520, but the remaining generated copies do not establish that each OS/architecture cell exercises its own platform behavior. Repeating a sizeable algorithm is not equivalent to covering different failure modes.

The intended division is: shared cases for language features/algorithms; shared/ for argument registers, stack spill, returns, alignment and integer extension; / for native file/memory/socket/event operations and their ABI, combined with real workloads.

Source evidence at the head of #520:

• tools/populate_case_matrix.py:95 — def generated_case
• tools/populate_case_matrix.py:96 — source = CASES_ROOT / "shared"

Acceptance:

• Audit the remaining generated cases and replace a coherent first group with distinct, deterministic, bounded-resource workloads at the proper layer.
• Preserve existing tests by strengthening or adding cases; do not delete coverage to reduce duplication.
• Use architecture-specific ABI pressure and OS operations with observable results, not assembly-only filler or unconditional success.
• Include varied valid syntax/layouts with readable multi-line code; keep tests/cases README concise rather than creating a case-by-case catalog.
• Keep generator refresh from restoring replaced copies, and record compile-only versus executed coverage accurately.

Related: #461 covers generator regression tests, #458 CPU profiles, and #459 shared portability. This issue concerns workload substance and division by layer.

Audit status: identified by static source inspection; the scenarios above have not been executed during this audit. The permalink fixes the reviewed revision; this report does not claim the defect was introduced by #520.