release blocker: argument-shape routes fall back to pshape_arg.fallback (issue_8774, shard 3) #9264

Symptom

cargo-test-perry (3/8) fails on release candidate 83754818ea (full-tier run 33372993990):

test guard_failures_match_node_and_unsafe_parameters_stay_generic ... FAILED
panicked at crates/perry/tests/issue_8774_argument_shape_clones.rs:359:5:
all selected semantic-fixture routes originate at fresh contained locals

The assertion is !ir.contains("pshape_arg.fallback") — the emitted IR now contains an argument-shape fallback block where the fixture expects every selected route to take its clone.

Reproduction

# MUST rebuild the static wrappers first — see "Environment" below.
cargo build -p perry -p perry-runtime-static -p perry-stdlib-static
RUST_TEST_THREADS=1 cargo test -p perry --test issue_8774_argument_shape_clones guard_failures

Takes ~55 s when it genuinely runs.

Attribution: UNKNOWN — and #9242 has been ruled out

It passed at candidate 01463b8161 and fails at 83754818ea. #9242 was the only commit touching lower_call/method_override.rs in that window, so it was the obvious suspect — and it is not the cause. Measured on a coherent tree, with #9242's class_chain_may_declare_method_field bail in three states:

Please do not re-attribute this to #9242 without new evidence; that hypothesis has been tested and rejected.

The cause is elsewhere in main between 01463b8161 and 83754818ea. A bisect over that window is the obvious next step — note it is a large window, and the fixture compiles three modules, so each step costs ~1 min plus build.

Acceptance criteria

• guard_failures_match_node_and_unsafe_parameters_stay_generic and stable_argument_clones_are_direct_reported_and_moving_gc_safe both pass.
• The cause is identified, not worked around by relaxing the assertion — it encodes the perf(codegen): propagate Ptr<Shape> facts into guarded argument clones #8774 invariant that selected routes originate at fresh contained locals.

Environment (this cost a full day elsewhere)

cargo test -p perry rebuilds the binary but not the runtime archives. After any git checkout/rebase, rebuild -p perry -p perry-runtime-static -p perry-stdlib-static or the compiler/runtime coherence check refuses every fixture with Error: runtime library does not match this Perry compiler — which surfaces as perry compile failed and a suite finishing in ~19 s instead of ~55 s. A fast suite is the tell.