chore(deps): switch mimalloc3 to upstream microsoft/mimalloc (#82)

### Background

On macOS, mimalloc v3 stores the per-thread heap pointer in a fixed TLS slot (TCB `[108]`/`[109]`) shared by every image, so a second statically-linked copy (e.g. another napi addon) adopts the first's heap and crashes. We worked around this with a vendored `prim.h` patch in the `napi-rs/mimalloc` fork. Upstream now fixes it in [microsoft/mimalloc#1301](microsoft/mimalloc#1301) ([b83dee64](microsoft/mimalloc@b83dee6)) by exposing cmake options, so we can drop the fork.

### Changes

- **`.gitmodules`** — `mimalloc3` url `napi-rs/mimalloc` → `microsoft/mimalloc`.
- **`mimalloc3`** — `5853e09` (fork) → `b83dee64` (official `dev3`); drops the vendored `prim.h` patch.
- **`build.rs`** — on v3 + macOS, enable the official options:
  ```rust
  if env::var_os("CARGO_FEATURE_V3").is_some() && target_os == "macos" {
      cmake_config
          .define("MI_TLS_MODEL_LOCAL", "ON")     // -> MI_TLS_MODEL_THREAD_LOCAL=1
          .define("MI_TLS_RECURSE_GUARD", "ON");  // -> MI_TLS_RECURSE_GUARD=1
  }
  ```
- **`mimalloc` (v2)** — routine bump `02a2f5d` → `fef6b0d`.

### Notes

- **Scope: v3 + macOS only.** Options exist only in v3; Linux already defaults to thread-local (no-op), Android/OpenBSD use pthreads, Windows uses its own TlsAlloc path. Add `target_os == "ios"` to cover iOS/tvOS.
- **`MI_TLS_RECURSE_GUARD` is belt-and-suspenders** — upstream `prim.h` already auto-enables it on Apple; we set it explicitly for clarity.
- **No regression of the dead-main-thread fix** (issue #1287 / rolldown#9722): the `MI_THREADID_INVALID` poison in `mi_tld_free` is present in `b83dee64`.

### Verification (macOS)

- v3 build: `CMakeCache` `MI_TLS_MODEL_LOCAL:BOOL=ON` / `MI_TLS_RECURSE_GUARD:BOOL=ON`; compiled with `MI_TLS_MODEL_THREAD_LOCAL=1` + `MI_TLS_RECURSE_GUARD=1`.
- Default (v2) build: options absent (gating works), builds clean.
- `cargo test --features v3`: `mimalloc-safe` 6/6 pass; `libmimalloc-sys-test` runs.

Author: shulaoda

.gitmodules

@@ -3,5 +3,5 @@
 	url = https://github.com/microsoft/mimalloc.git
 [submodule "libmimalloc-sys/c_src/mimalloc3"]
 	path = libmimalloc-sys/c_src/mimalloc3
-	url = https://github.com/napi-rs/mimalloc.git
+	url = https://github.com/microsoft/mimalloc.git
 	branch = dev3

libmimalloc-sys/build.rs

@@ -76,10 +76,18 @@ fn main() {
         cmake_config.define("MI_LOCAL_DYNAMIC_TLS", "ON");
     }
 
-    // Note: the previous `-DMI_HAS_TLS_SLOT=0` workaround for Apple has been
-    // replaced by a vendored source patch in our mimalloc fork (napi-rs/mimalloc
-    // dev3) that routes Apple to `MI_TLS_MODEL_THREAD_LOCAL + MI_TLS_RECURSE_GUARD`
-    // directly in `include/mimalloc/prim.h`. See that patch for the full rationale.
+    // On macOS, mimalloc v3 defaults to a fixed TLS slot (TCB[108]/[109]) shared by
+    // every image, so a second statically-linked copy (e.g. another napi addon)
+    // adopts the first's per-thread heap and crashes. Force the thread-local model
+    // so each image gets its own `__thread` storage (a v3-only cmake option from
+    // microsoft/mimalloc b83dee64, issue #1301).
+    if env::var_os("CARGO_FEATURE_V3").is_some() && target_os == "macos" {
+        cmake_config
+            .define("MI_TLS_MODEL_LOCAL", "ON")
+            // macOS may allocate on a thread-local's first access; guard against
+            // re-entering malloc until the process is initialized.
+            .define("MI_TLS_RECURSE_GUARD", "ON");
+    }
 
     if (target_os == "linux" || target_os == "android")
         && env::var_os("CARGO_FEATURE_NO_THP").is_some()