rebless: state_mutation contract — writes, not declarations (gitgalaxy#2765, Phase 3)

[squid-protocol]

Re-bless for gitgalaxy#2765 (the state_mutation contract, roadmap #2812 Phase 3). Engine PR: squid-protocol/gitgalaxy#2818.

What the contract says

One hit is a statement that writes a new value into state that already exists. A declaration is not a write (even var/let mut/int x = 5/go :=); a type or modifier naming mutable state is not a write; a read, bind or cast is not a write; one statement is one hit and a token another rule owns is not a second signal. docs/state_mutation_rule_contract.md in gitgalaxy has the corollaries and the 46-language audit.

What moved here, and why (every cell is in the ledger entry state-mutation-contract-2765)

Cells that moved because the rule was wrong (no plant change): haskell a.hs 2->0 (IORef type signatures), dockerfile a 2->0 (ENV is globals), c a 2->0 / b 3->2 (declarations), go b 5->2 (:=, _ = x), objective-c b 3->2, perl b 3->2 / c 1->0 (my, bare shift), m4 c 2->0 (popdef is cleanup), apex c 1->0 (.clear( is cleanup), matlab c 1->0 and b cleanup 2->0 (clear/clearvars are cleanup only -- matlab's corpus cleanup now sits on the median).

Plants authored to the old rules, re-planted as real writes: cpp, java, javascript, kotlin, rust, scala, swift, typescript, zig (var first = items x2 -> one declaration + first = 1 / first = 2), solidity (payable(items) is a cast -> items = 1), sqlite (UPDATE ... SET had counted twice -> ALTER TABLE corpus RENAME TO corpus_state is the second statement; io stays 3), abap (the = statement joined the rule, APPEND dropped), livecode/scheme/tcl (the decoy line was already a third write; one plant removed). The probe_debt bodies in abap/fortran/matlab/powershell assigned into languages with no declaration syntax and carried a state_mutation into a c file planted at 0; they declare or return now. Every replacement was screened as a pair against the branch rules; the only unplanted side effect is java's AtomicInteger no longer carrying a concurrency 1.

Ledger: new state-mutation-contract-2765 (upstream-bug, resolved, verifying scan 46/46 against fix/2765-state-mutation-contract); batch4-dual-keyword-overlaps loses matlab and m4 (their only itemised duals were retired) and records jcl SET as the one state_mutation dual that stays. tools/screen_plant.py now honours $GITGALAXY_PATH.

Still narrower than the contract, by design for now: python/embedded_python/ruby never count a plain local re-assignment -- gitgalaxy#2817, not a red cell.

Effect (regenerated at full precision against the branch)

state_mutation reads 2 in all 43 comparable languages -- 0 out-of-band cells, down from 16 open defects (37%); risk_state_flux open-defect share 0 of 45. Corpus open-defect share 79 -> 62 of 2303 cells (3.4% -> 2.7%); average consistency across 49 metrics 88% -> 90%. decoy_check.py --gate and na_check.py clean.

Cross-repo

gitgalaxy#2818 merges first; verify.yml here checks out engine main, so this PR is red until then and green by construction after. bias-history.yml regenerates the report and chart on merge (the locally regenerated artifacts are included so the effect is visible in the diff).