Associate each thread with a model (#29573)

This PR makes it possible to use different LLM models in the agent
panels of two different projects, simultaneously. It also properly
restores a thread's original model when restoring it from the history,
rather than having it use the default model. As before, newly-created
threads will use the current default model.

Release Notes:

- Enabled different project windows to use different models in the agent
panel
- Enhanced the agent panel so that when revisiting old threads, their
original model will be used.

---------

Co-authored-by: Richard Feldman <oss@rtfeldman.com>

crates/agent/src/tool_use.rs

@@ -7,7 +7,7 @@ use futures::FutureExt as _;
 use futures::future::Shared;
 use gpui::{App, Entity, SharedString, Task};
 use language_model::{
-    LanguageModel, LanguageModelRegistry, LanguageModelRequestMessage, LanguageModelToolResult,
+    ConfiguredModel, LanguageModel, LanguageModelRequestMessage, LanguageModelToolResult,
     LanguageModelToolUse, LanguageModelToolUseId, MessageContent, Role,
 };
 use ui::IconName;
@@ -353,7 +353,7 @@ impl ToolUseState {
         tool_use_id: LanguageModelToolUseId,
         tool_name: Arc<str>,
         output: Result<String>,
-        cx: &App,
+        configured_model: Option<&ConfiguredModel>,
     ) -> Option<PendingToolUse> {
         let metadata = self.tool_use_metadata_by_id.remove(&tool_use_id);
 
@@ -373,13 +373,10 @@ impl ToolUseState {
 
         match output {
             Ok(tool_result) => {
-                let model_registry = LanguageModelRegistry::read_global(cx);
-
                 const BYTES_PER_TOKEN_ESTIMATE: usize = 3;
 
                 // Protect from clearly large output
-                let tool_output_limit = model_registry
-                    .default_model()
+                let tool_output_limit = configured_model
                     .map(|model| model.model.max_token_count() * BYTES_PER_TOKEN_ESTIMATE)
                     .unwrap_or(usize::MAX);