Standardize on u64 for token counts (#32869)

Previously we were using a mix of `u32` and `usize`, e.g. `max_tokens:
usize, max_output_tokens: Option<u32>` in the same `struct`.

Although [tiktoken](https://github.com/openai/tiktoken) uses `usize`,
token counts should be consistent across targets (e.g. the same model
doesn't suddenly get a smaller context window if you're compiling for
wasm32), and these token counts could end up getting serialized using a
binary protocol, so `usize` is not the right choice for token counts.

I chose to standardize on `u64` over `u32` because we don't store many
of them (so the extra size should be insignificant) and future models
may exceed `u32::MAX` tokens.

Release Notes:

- N/A

crates/deepseek/src/deepseek.rs

@@ -58,8 +58,8 @@ pub enum Model {
         name: String,
         /// The name displayed in the UI, such as in the assistant panel model dropdown menu.
         display_name: Option<String>,
-        max_tokens: usize,
-        max_output_tokens: Option<u32>,
+        max_tokens: u64,
+        max_output_tokens: Option<u64>,
     },
 }
 
@@ -94,14 +94,14 @@ impl Model {
         }
     }
 
-    pub fn max_token_count(&self) -> usize {
+    pub fn max_token_count(&self) -> u64 {
         match self {
             Self::Chat | Self::Reasoner => 64_000,
             Self::Custom { max_tokens, .. } => *max_tokens,
         }
     }
 
-    pub fn max_output_tokens(&self) -> Option<u32> {
+    pub fn max_output_tokens(&self) -> Option<u64> {
         match self {
             Self::Chat => Some(8_192),
             Self::Reasoner => Some(8_192),
@@ -118,7 +118,7 @@ pub struct Request {
     pub messages: Vec<RequestMessage>,
     pub stream: bool,
     #[serde(default, skip_serializing_if = "Option::is_none")]
-    pub max_tokens: Option<u32>,
+    pub max_tokens: Option<u64>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub temperature: Option<f32>,
     #[serde(default, skip_serializing_if = "Option::is_none")]