feat(server): use native tokenizer impl (#6960)

### Benchmark

`yarn workspace @affine/server-native bench`

```
┌─────────┬────────────┬─────────┬────────────────────┬──────────┬─────────┐
│ (index) │ Task Name  │ ops/sec │ Average Time (ns)  │ Margin   │ Samples │
├─────────┼────────────┼─────────┼────────────────────┼──────────┼─────────┤
│ 0       │ 'tiktoken' │ '5'     │ 176932518.76000002 │ '±4.71%' │ 100     │
│ 1       │ 'native'   │ '16'    │ 61041597.51000003  │ '±0.60%' │ 100     │
└─────────┴────────────┴─────────┴────────────────────┴──────────┴─────────┘
```

packages/backend/native/src/lib.rs

@@ -2,12 +2,17 @@
 
 pub mod file_type;
 pub mod hashcash;
+pub mod tiktoken;
 
 use std::fmt::{Debug, Display};
 
 use napi::{bindgen_prelude::*, Error, Result, Status};
 use y_octo::Doc;
 
+#[cfg(not(target_arch = "arm"))]
+#[global_allocator]
+static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
 #[macro_use]
 extern crate napi_derive;
 

packages/backend/native/src/tiktoken.rs

@@ -0,0 +1,30 @@
+use std::collections::HashSet;
+
+#[napi]
+pub struct Tokenizer {
+  inner: tiktoken_rs::CoreBPE,
+}
+
+#[napi]
+pub fn from_model_name(model_name: String) -> Option<Tokenizer> {
+  let bpe = tiktoken_rs::get_bpe_from_model(&model_name).ok()?;
+  Some(Tokenizer { inner: bpe })
+}
+
+#[napi]
+impl Tokenizer {
+  #[napi]
+  pub fn count(&self, content: String, allowed_special: Option<Vec<String>>) -> u32 {
+    self
+      .inner
+      .encode(
+        &content,
+        if let Some(allowed_special) = &allowed_special {
+          HashSet::from_iter(allowed_special.iter().map(|s| s.as_str()))
+        } else {
+          Default::default()
+        },
+      )
+      .len() as u32
+  }
+}