Bound P104 endpoint support and reject the 406-lifetime profile

[petalas]

Summary

Close the compatible-endpoint handoff from #2889 for its exact 406-lifetime profile.

335 histories (1,7,1) + 71 unit-step histories
  endpoint-span total       832 >= 832       passes the earlier screen
  endpoint-support total   2465 < 4656       cannot realize the correction

Adds only research/p104-endpoint-support-capacity/: a scoped proof, two separately implemented geometric checkers, source pins, receipt, tests and registered replay. Each orientation has 4,656 residual demands; every proper endpoint subspace carries at most 30 residual labels. The resulting necessary rows cover all 256 complete dimension histories.

Credit Hiroki Kumazaki/#2889/#2353 for the predecessor and handoff, ywz65/#2614, Arham Tahir/#2675/#2682, Sagar Thakrar's mixer handoff, Philippe Van Velzen, Douglas Colkitt and the retained geometry/native lineage. Exact heads and credit are in the package. Canonical claim and handoff.

Evidence

• Before: Audit P104 helper histories with endpoint span constraints #2889's 406 profile passes its span rows and declared favorable cost screen. After: the new support rows reject it, with a 2,191-demand deficit per orientation.
• Direct hyperplane enumeration and independent bitset/Walsh-Hadamard calculations agree on all 208 endpoint records.
• A different 400-lifetime profile still passes the new rows and the inherited exact moment relaxation, upper bound 0.999729278053145. No route-wide rejection follows.
• CPython 3.12.15 registered replay and five behavioral test groups pass, including source/receipt faults, semantic mutants and optimized read-only Git-free execution. CPython 3.13.5 direct checks, 133 research-workflow tests, Ruff, warning-as-error compilation, selected-result binding and committed-byte preflight pass.

python3.12 scripts/research_package.py --package research/p104-endpoint-support-capacity

Started at main 2b4c46fbb; publication fast-forwarded to a1bfbc677. Relevant guidance, priority revision 5, selected record, scientific inputs and #2889 head were unchanged. Final overlap covered 488 open PRs, 2,103 closed PR bodies and 1,558 Discussion records with complete pagination; arbitrary diffs and unpublished work are outside coverage.

Merge Danger

Door: two-way. One new opt-in research package; no predecessor or selected-frontier changes.

Blast Radius: scoped. This is a written necessary-condition proof with finite premise checks, not a correction circuit or full multiplication theorem. Physical/native execution, chronology, restoration, hosting, precision, complete consumer costs and all-size admission remain open. No new supplier or kappa, independent human/formal review, or hosted-CI success is claimed.