Run frontend driver regressions in isolated LLVM backend CI jobs

## Gap

The frontend driver tests introduced in #486 select an enabled LLVM target, so they can run in isolated backend builds. However, `.github/workflows/rust.yml` currently runs the isolated WebAssembly and LoongArch jobs with only `--test codegen_regressions` and a name filter (`webassembly_` or `loongarch64_`). Those steps never execute `tests/frontend_regressions.rs`. The default workspace and full core64 lanes cover these tests, but cannot catch accidental reliance on another enabled backend in the isolated configurations.

This is an additional CI coverage task after #486, not a request to repair a currently failing frontend test.

## Small contribution

Add a separate frontend driver step to each existing isolated WebAssembly and LoongArch job:

```bash
cargo test --locked --no-default-features --features llvm-target-wasm --test frontend_regressions --jobs 2
cargo test --locked --no-default-features --features llvm-target-loongarch --test frontend_regressions --jobs 2
```

Run each command in its corresponding existing job after toolchain/stdlib setup. Do not pass the codegen test-name filter to these commands: it would select zero frontend tests. No new runner, backend implementation, or target matrix is needed.

On #486's local head `6fb6ed7`, WebAssembly frontend tests passed 4/4 and the LoongArch frontend suite passed 3/3 during validation.

## Acceptance

- [ ] Both isolated jobs execute the frontend test binary and report a nonzero test count.
- [ ] Existing codegen/ABI smoke steps remain intact.
- [ ] The WebAssembly suite exercises wasm32 versus wasm64 pointer-sized integer behavior.
- [ ] Commands retain `--locked` and `--jobs 2`.

Start in `.github/workflows/rust.yml` and `tests/frontend_regressions.rs`. This is suitable for a first contribution because the tests and required toolchains already exist.

Start in .github/workflows/rust.yml and compare the existing isolated WebAssembly and LoongArch jobs, then inspect tests/frontend_regressions.rs to confirm the frontend test binary and target coverage. Add the separate commands after toolchain and stdlib setup while keeping the existing codegen/ABI steps intact. Done means both jobs report a nonzero frontend test count with --locked and --jobs 2.

Written by the indexing model from the issue text.