Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@tests/test_codegen.py`:
- Around line 10468-10536: Add a regression test method to
TestGCShadowStackOverflow that deliberately exceeds the shadow-stack guard and
asserts the runtime traps by calling _run_trap(...) on a program invoking
build_acc (or equivalent recursive main) with a very large depth (e.g., far
above the guarded limit); name it something like
test_deep_array_accumulation_trap and have it invoke the same build_acc/main
pattern used in existing tests but with a depth large enough to trigger the
overflow guard so the test fails with a trap instead of returning.

In `@vera/codegen/assembly.py`:
- Around line 508-514: In _emit_gc_collect where the Phase 2 worklist pushes use
wl_base/wl_end and currently skip enqueuing on overflow, replace the silent-skip
behavior with a fail-fast check: before incrementing or storing via wl_ptr,
compare wl_ptr and wl_end (or compute next_ptr) and if the worklist is full emit
a trap/abort or call the runtime panic path so the collector fails immediately
instead of dropping marks; update both push sites that reference wl_ptr/wl_end
to perform this check and call the same failure routine (or wasm unreachable) to
make overflow deterministic.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@ROADMAP.md`:
- Line 229: Update the completed-phases summary in ROADMAP.md where it currently
reads "**630+ commits, 111 tagged releases, 3,253 tests, 96% coverage, 73
conformance programs, 30 examples, 13 spec chapters.**" to reflect the new
release count by changing "111 tagged releases" to "112 tagged releases"
(matching the 0.0.112 PR); ensure the single-line summary containing that exact
phrase is edited so other docs remain consistent.

In `@tests/test_codegen.py`:
- Around line 1657-1658: The assertions are too loose and can match the same
numeric constant used for `$gc_heap_start`; update the checks to bind the
constant to the `$heap_ptr` global specifically by asserting the initializer of
`global $heap_ptr` contains `i32.const 32773` (e.g. use a regex or parse
`result.wat` to find the `global $heap_ptr` line and verify it includes
`i32.const 32773`), and apply the same tighter check for the other test case
that currently checks `i32.const 32768` so both constants are validated only as
`$heap_ptr` initializers.
- Around line 10542-10563: The test
test_deep_array_accumulation_preserves_content only asserts
array_length(build_acc([], 500)) == 500 and therefore doesn't validate element
values; update the test to actually check contents produced by build_acc (or
rename it to indicate it's length-only). Specifically, in the test referencing
build_acc and main/array_length, either (preferred) iterate over build_acc([],
500) and assert every element is false (or compare to an expected array of 500
falses) to ensure values weren't corrupted, or change the test name to
test_deep_array_accumulation_preserves_length and leave only the length
assertion.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@ROADMAP.md`:
- Line 11: Update the three occurrences of the test total that currently read
"3,253 tests" to "3,252 tests" so the document matches the PR's reported results
(3,250 → 3,252 and 3,241 passed + 11 skipped = 3,252); locate and replace the
exact sentence fragment "The compiler is complete end-to-end: ... The compiler
has 3,253 tests" (and the two other identical lines noted in the review) so all
three instances show "3,252" consistently.

---

Duplicate comments:
In `@tests/test_codegen.py`:
- Around line 10480-10513: The test only checks index 0 via first_bool(main), so
tighten it to verify a prefix (e.g., indices 0..3) to catch corruption of the
first few elements: update the Wasm source in test_deep_array_accumulation_bool
(functions build_acc, first_bool, public fn main) so that main inspects at least
elements 0 through 3 of the result of build_acc (returning 0 only if all four
are false, 1 otherwise), then keep the Python assertion assert _run(src) == 0;
this ensures the test fails if any of the first four elements are corrupted.