fix: x86 SMP bitcode/code-model mismatch (smp_semaphore/mutex/threads)

[avrabe]

Summary

Three CI checks have been failing for at least the last several PRs:

• `smp_semaphore (qemu_x86_64, SMP)`
• `smp_mutex (qemu_x86_64, SMP)`
• `smp_threads (qemu_x86_64, SMP)`

All three fail at the same step — building `libgale_ffi.a` for `x86_64-unknown-none`:

```
error: failed to load bitcode of module
"core-fcc82a7537a7267a.core.bb170dea1013a3a3-cgu.0.rcgu.o"
error: could not compile gale-ffi (lib) due to 1 previous error
```

Root cause

Reproduced locally with rustc 1.95.0. The CI log truncates one line earlier than the actual cause; the smoking gun is a warning that precedes the error:

```
warning: linking module flags 'Code Model':
IDs have conflicting values: 'i32 2' from , and 'i32 1' from gale_ffi.5b37a6aa01c132b8-cgu.0
```

The chain:

• `gale/ffi/Cargo.toml` sets `lto = true` for `[profile.release]` (fat LTO).
• `gale/zephyr/CMakeLists.txt` sets `-Ccode-model=small` for x86 (correct — Zephyr maps the kernel low).
• The precompiled `core` rustlib for `x86_64-unknown-none` was built with the target's default `code-model=kernel` and ships with embedded bitcode tagged that way.
• rustc fat-LTO loads `core`'s bitcode and tries to merge it with `gale-ffi`'s bitcode. LLVM's "Code Model" module flag is `Error`-level: 1 (small) and 2 (kernel) cannot coexist, so the bitcode load aborts.

Cortex-M doesn't trip on this because we don't override `code-model` on Cortex-M; the default matches the precompiled `core`.

Fix

Switch x86 builds from the `release` profile to the existing `release-lto` profile (which has `lto = false`), but only when the cross-language-LTO upper block hasn't already set it. `release-lto` already `inherits = "release"` so all other tuning (`opt-level="z"`, `codegen-units=1`, `panic="abort"`, `overflow-checks=true`) is preserved.

Scoped inside `if(CONFIG_X86)` so Cortex-M is byte-identical to before — the merged-in PR #27 publication-relevant LTO numbers are not at risk.

```diff
diff --git a/zephyr/CMakeLists.txt b/zephyr/CMakeLists.txt
@@ -183,6 +183,29 @@ if(CONFIG_X86)
...

• if(GALE_CARGO_PROFILE_DIR STREQUAL "release")
• set(GALE_CARGO_PROFILE_ARGS "--profile" "release-lto")
• set(GALE_CARGO_PROFILE_DIR "release-lto")
• message(STATUS "Gale: x86 — using release-lto profile to avoid core bitcode/code-model mismatch")
• endif()
  endif()
  ```

Test plan

• Reproduced the failure locally (rustc 1.95.0, x86_64-unknown-none target, with the exact CI feature set)
• Confirmed fix resolves locally — same cargo invocation builds clean producing 10 MB libgale_ffi.a
• Cortex-M cargo build (thumbv7m-none-eabi) still builds clean — change is `if(CONFIG_X86)`-scoped
• CI on this PR validates the three SMP tests now pass

Local repro / validation command (no Zephyr SDK needed):

```
cd gale/ffi && env RUSTFLAGS="-Crelocation-model=static -Ccode-model=small" \
GALE_MAX_SEMS=32 GALE_MAX_WAITERS=16 \
cargo build --profile release-lto --target x86_64-unknown-none \
--no-default-features \
--features sem,mutex,condvar,msgq,pipe,stack,timer,...,rbtree
```

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!