test: port upstream x86 unit tests

[KeyCode17]

x86 is the only architecture whose upstream unit-test file has no Rust counterpart, even though it is by far the largest one. Every other arch was ported:

upstream tests/unit/	size	Rust module
test_x86.c	71 KB	(none)
test_arm64.c	34 KB	tests/arm64.rs
test_arm.c	33 KB	tests/arm.rs
test_riscv.c	25 KB	tests/riscv.rs
test_mem.c	19 KB	tests/mem.rs
test_ctl.c	12 KB	tests/ctl.rs
test_mips / ppc / m68k / s390x	1–8 KB	all present

x86 appears incidentally in ctl.rs and mem.rs, but nothing covers the x86-specific behaviour that test_x86.c exercises: SMC, MMIO, TCG-op hooks, nested emu_start, TB-cache invalidation, x87 fnstenv, fxsave FPIP, bswap/REX prefix handling, unaligned access logging.

What's in it

• crates/unicorn/src/tests/x86.rs — 50 tests ported from tests/unit/test_x86.c at the current submodule pin (681bd32), following the existing per-arch module conventions (uc_common_setup, state via new_with_data + get_data_mut, assert_eq! over TEST_CHECK).
• crates/unicorn/src/tests/mod.rs — registers the module behind #[cfg(feature = "arch_x86")], matching the other arch gates.

No library code is touched.

Not ported (6 of 56), and why

These need bindings that don't exist yet. I deliberately left them out rather than adding API in a test PR — happy to follow up with the bindings if you want them.

upstream test	missing binding
test_x86_hook_cpuid, test_x86_hook_insn_rdtsc, test_x86_hook_insn_rdtscp	an instruction hook whose callback returns a value (uc_cb_insn_cpuid_t). add_insn_sys_hook takes FnMut(..) -> (), so the "skip the instruction" return can't be expressed.
test_x86_mmu, test_x86_read_virtual	uc_x86_msr read/write. MSR isn't in value_size_x86, and reg_read only yields u64.
test_x86_segmentation	a uc_x86_mmr accessor to write GDTR.

One judgement call worth flagging

test_x86_unaligned_access / test_x86_64_unaligned_access sit behind #if !defined(TARGET_READ_INLINED) && defined(BOOST_LITTLE_ENDIAN) upstream (CMake sets TARGET_READ_INLINED for aarch64 and ppc). I mirrored that guard with a cfg. Both tests actually pass on aarch64 here when I remove the guard — so it can be dropped if you'd rather have the coverage on Apple Silicon / arm64 CI. I kept it to match upstream's stated constraint rather than silently diverging.

Verification

• cargo test — 154 passed, 0 failed (106 before, +48 on this aarch64 host; the two guarded tests bring it to +50 on x86_64).
• cargo fmt --all --check — clean.
• cargo clippy --workspace --all-targets — no new hits from this file. (There are 3 pre-existing -D warnings errors in crates/unicorn-sys/build.rs under clippy 1.96 — collapsible_if ×2 and unnecessary_debug_formatting. Untouched here; can send a separate PR if you'd like.)
• Spot-checked that the hook-driven tests aren't passing vacuously: probed that the fnstenv and mem_hooks_pc_guarantee callbacks actually fire, and confirmed a deliberately broken expectation does fail.

[PhilippTakacs]

I don't think there is any benefit in copy the unit test from unicorn. The rust test should test the rust api. If unicorn is working correct is tested by unicorn.

[wtdcode]

Same feeling. This does not make too much sense and it will be out of sync easily.