Research: audit later-stage selected BIT reserve composition

[petalas]

Summary

Add research/selected-bootstrap-reserve-composition/, an independent conditional audit and regression for later selected BIT bootstrap stages.

previous callee completed at every width and every child return
  implies reusable external stock through stages 2 through 7
whole two-call endpoint alone
  does not imply the required child-return contract

The source-contract induction corroborates the existing reserve-composition argument. It inherits the conditional coefficient 6748 from #2120; it does not derive a new stock formula or admit an actual all-size implementation. The new finite regression distinguishes the individual return contract from an apparently correct completed wrapper endpoint.

Card selected-stock-envelope, priority revision 4, refined to the later-stage completion implication rather than repeating first-stage headroom arithmetic. Scientific base is 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2.

Credit Rohan Arun / retained PR185 and PR387 for nested restoration, Adam Gall / #2084 at 8001a23b003bec5dff5e61edec907b652c475932 for the conditional selected BIT implication, and Nick Petalas / #2120 at 378593574261e0b0915e7e8a83147aa640274fc8 for its coefficient refinement. Josie Elliston / #1956 and #1965, ToikY / #1990, and the selected supplier lineage retain credit. #2399 and #2457 address complementary callee/spectator obligations and are not admitted by this audit.

Tested artifact

Evidence

• Before: two outer-row negations cancel at the final endpoint in an F3, stage-two, width-four wrapper. An endpoint-only check misses the intermediate row debt.
  After: the individual child-return regression rejects the exact trajectory 1, 2, 1, while valid nested wrappers pass. This is a countermodel to a weakened premise, not a selected-implementation bug report.
• The replay checks 144 complete F3/F5 maps over 1872 basis columns, all fourteen source-bound fit guards, source/archive-member pins and eight tests with semantic, capacity and receipt failure controls.

python3.12 -Werror -B scripts/research_package.py \
  --package research/selected-bootstrap-reserve-composition

Checked with CPython 3.12.15 on Linux x86_64 / WSL2. Actual all-width packet/compiler/callee completion, complete-spectator conversion, H2 uniform time, preparation, parking and precision remain assumptions. No full selected cold replay, new exponent or supplier admission is claimed.

Merge Danger

Door: two-way

Blast Radius: research

Only this opt-in package and its byte-pinned predecessor text are added. Selected records and frozen source inputs are unchanged. The composition conclusion is conditional, and the finite fixture is not a q-adic tape implementation.