Use u64 instead of usize for durations, counts, and sizes

usize is platform-dependent (32-bit on 32-bit targets). Nanosecond
durations can exceed u32::MAX in ~4.3 seconds, and model sizes in
bytes can easily exceed 4 GiB. Using u64 ensures correctness across
all platforms.

Changed fields:
- GenerateResponse: total_duration, load_duration, prompt_eval_count,
  prompt_eval_duration, eval_count, eval_duration
- Model: size
- RunningModel: size, size_vram

src/types/generate.rs

@@ -137,20 +137,20 @@ pub struct GenerateResponse {
     pub done_reason: Option<String>,
 
     /// Time spent generating the response in nanoseconds
-    pub total_duration: Option<usize>,
+    pub total_duration: Option<u64>,
 
     /// Time spent loading the model in nanoseconds
-    pub load_duration: Option<usize>,
+    pub load_duration: Option<u64>,
 
     /// Number of input tokens in the prompt
-    pub prompt_eval_count: Option<usize>,
+    pub prompt_eval_count: Option<u64>,
 
     /// Time spent evaluating the prompt in nanoseconds
-    pub prompt_eval_duration: Option<usize>,
+    pub prompt_eval_duration: Option<u64>,
 
     /// Number of output tokens generated in the response
-    pub eval_count: Option<usize>,
+    pub eval_count: Option<u64>,
 
     /// Time spent generating tokens in nanoseconds
-    pub eval_duration: Option<usize>,
+    pub eval_duration: Option<u64>,
 }

src/types/ps.rs

@@ -11,10 +11,10 @@ pub struct PsResponse {
 pub struct RunningModel {
     pub name: String,
     pub model: String,
-    pub size: usize,
+    pub size: u64,
     pub digest: String,
     pub details: ModelDetails,
     pub expires_at: String,
-    pub size_vram: usize,
+    pub size_vram: u64,
     pub context_length: u32,
 }

src/types/tags.rs

@@ -12,7 +12,7 @@ pub struct Model {
     pub name: String,
     pub model: String,
     pub modified_at: String,
-    pub size: usize,
+    pub size: u64,
     pub digest: String,
     pub details: ModelDetails,
 }