t27c gen: strength_reduce turns checked integer x * 2^k into wrapping x << k #6290

Refs #6063

strength_reduce / reduce_expr in bootstrap/src/compiler.rs rewrites x * 2^k as x << k before Zig gen. For integers this changes semantics:

• Zig's * is checked and panics on overflow.
• Zig's << drops bits silently.

#6284 covers the f64 case only. This is the integer case.

Repro (t27c test-report, /tmp/t27shared/release/t27c):

module sr;

fn f(x: u8) u8 {
    return x * 128;
}

test mul_overflows {
    assert(f(3) == 128);
}

• t27c gen emits return x << 7;.
• t27c test-report reports 1 pass, because 3 * 128 wraps to 128.
• With no rewrite, Zig would panic with integer overflow.
• t27b (the native backend, checked *) traps integer overflow at line 4 (* on u8).

The reference and t27b therefore disagree on any spec that multiplies by a power of two and overflows. Expected: keep * (or emit std.math.shl with an overflow check) whenever overflow is possible.

Found while doing t27b shifts (#6288, PR #6289).

Boundary

• bootstrap/src/compiler.rs

User Scenarios

• Given a t27 spec with a function that multiplies a u8 by 128 (2^7), When t27c gen compiles it, Then the emitted Zig code must use checked multiplication (*) or a checked shift, not an unchecked left shift (<<), so that overflow panics instead of wrapping.
• Given the test module from the repro, When running t27c test-report, Then the test must fail (exit non-zero) because the checked multiplication of 3 * 128 overflows u8, matching t27b's trap behavior.

Requirements

• FR-001: The strength_reduce / reduce_expr pass in bootstrap/src/compiler.rs MUST NOT rewrite integer x * 2^k into x << k when the multiplication is checked and overflow is possible.
• FR-002: For integer types, the compiler MUST preserve checked multiplication semantics (panic on overflow) unless it can prove the operation cannot overflow.
• FR-003: If a shift is emitted for multiplication by a power of two, it MUST be a checked shift (e.g., std.math.shl with overflow check) that panics on overflow, matching Zig's * behavior.

Success Criteria

• t27c gen on the repro module emits return x * 128; (or a checked shift equivalent) and not return x << 7; — verified by inspecting bootstrap/src/compiler.rs output.
• Running t27c test-report on the repro module exits non-zero (test fails due to overflow panic) — the command t27c test-report exits with non-zero.
• The existing test suite passes (no regressions) — cargo test in the bootstrap crate exits 0.