perf: quantify memory costs and broaden call-path evidence after #1870 #1885

The measured throughput acceptance criteria from #1863 are now met by the candidate in PR #1870. The exact pinned-blead 14-pair portfolio reports PerlOnJava/Perl geometric means of 1.059x for the method anchor (95% interval 1.040–1.088), 1.547x for closure (1.524–1.564), and 1.100x for bit-packed Life (1.058–1.167). The report is dev/bench/reports/issue-1863-pr-port-full-14pair-blead.json.

The broader memory and call-path investigation in #1863 is not complete. Track those remaining items here:

• Compare allocation per operation, post-GC live heap, GC cost, and peak RSS against exact origin/master under the same fixed heap and workload.
• Profile ordinary direct calls, coderef calls, and method calls outside the stable-hash acceptance shape; identify inclusive/exclusive dispatch, frame, argument, and result costs.
• Verify retained argument-array pooling stays bounded and does not retain escaped values under repeated stress.
• Preserve reproducible reports and environment manifests for each comparison.

This follow-up does not change the measured 1:1 result for the three defined throughput workloads. It tracks the remaining memory and generality work separately.