w6c+selfhost: localloadop helper for sign-aware ident loads (closes #19)

Read-side fix dual to fldloadop: signed-narrow local/global ident loads
now MOVSXD/MOVSWQ/MOVSBQ from the slot instead of raw MOVQ. Deref-stores
(MOVL/MOVW/MOVB) no longer corrupt downstream i64 widens. Compound RMW
restructured to gate direct-mem ADDQ/SUBQ on load_op == MOVQ. Top-level
lets use LEAQ+indirect (w6a doesn't expose MOVSXD/MOVSWQ/MOVSBQ for
D_EXTERN).

dotchainresolve out-params restored to natural *i32 (workaround retired).
selfhost/CLAUDE.md graduated.

selfhost/CLAUDE.md

@@ -8,8 +8,6 @@ The C bootstrap's cgen has known silent-miscompilation traps. They produce wrong
 
 1. **`amalloc(n)` with n < struct size silently corrupts neighbours.** No error — the bump arena hands out n bytes and field writes overflow into the next record. When introducing or growing a struct, audit every `amalloc(_, n)` call site and over-size (we routinely pass 48 for a 40-byte struct). Symptom: linked-list prepends lose all but the most recent entry.
 
-2. **Out-param `*p: *i32` deref-stores leave caller slot's upper 4 bytes stale.** `*p = v` lowers to MOVL (4B), but the caller's 8B slot was zero-init via MOVQ; a later i64-widening read emits MOVQ (8B raw) and zero-extends, so a negative i32 round-trips as a 4G-rooted positive i64. Default to `*i64` out-params for offsets / signed indices until task #19 lands. Symptom: garbage frame offsets like `MOVL AX, 4294967280(BP)` in stage-2 asm.
-
 Fixed (no workaround needed):
 
 - `def NAME: str = "..."` field access. `.len`/`.ptr` on an Sdef ident
@@ -74,5 +72,22 @@ Fixed (no workaround needed):
   follow the per-index-write contract and stay uninit. See
   cmd/w6c/cgen.c N_LET (else branch, sz > 8 && !TY_ARRAY) and
   selfhost/cmd/wcc/cgenstmt.ww cglet no-rhs branch.
+- Signed-narrow scalar reads sign-extend honestly. `*p: *i32 = v`
+  stores 4 bytes (MOVL — correct for a 4B pointee), and the read
+  side now emits MOVSXD/MOVSWQ/MOVSBQ for any signed-narrow local
+  or let load instead of a raw MOVQ. The natural N_ASSIGN / N_LET
+  store paths happened to leave the slot sign-extended so MOVQ
+  reads accidentally worked; deref-stores were the only path that
+  touched fewer bytes than MOVQ reads, so negative i32 values
+  round-tripped as 4G-rooted positive i64s. Cgen now routes
+  scalar reads through `localloadop(t)`; aliases (`type err = !i32`)
+  resolve the size through TBANG / TNAME / TENUM. See cmd/w6c/cgen.c
+  N_IDENT/N_DOT mod.name catch-alls, cgassign compound RMW for
+  locals + top-level lets, and the matching selfhost cgenexpr.ww
+  cgident / cgassign / module-qualified branches plus
+  cgenutil.ww `localloadop`. Top-level lets and `mod.name` go
+  through a LEAQ + indirect load since w6a has no MOVSXD/MOVSWQ/
+  MOVSBQ D_EXTERN form. dotchainresolve's out-params are back at
+  their natural i32 width.
 
 If a port "should work" but the binary is wrong, suspect these first.