Make LanguageModel::use_any_tool return a stream of chunks (#16262)

This PR is a refactor to pave the way for allowing the user to view and
edit workflow step resolutions. I've made tool calls work more like
normal streaming completions for all providers. The `use_any_tool`
method returns a stream of strings (which contain chunks of JSON). I've
also done some minor cleanup of language model providers in general,
removing the duplication around handling streaming responses.

Release Notes:

- N/A

crates/language_model/src/language_model.rs

@@ -8,7 +8,7 @@ pub mod settings;
 
 use anyhow::Result;
 use client::{Client, UserStore};
-use futures::{future::BoxFuture, stream::BoxStream};
+use futures::{future::BoxFuture, stream::BoxStream, TryStreamExt as _};
 use gpui::{
     AnyElement, AnyView, AppContext, AsyncAppContext, Model, SharedString, Task, WindowContext,
 };
@@ -76,7 +76,7 @@ pub trait LanguageModel: Send + Sync {
         description: String,
         schema: serde_json::Value,
         cx: &AsyncAppContext,
-    ) -> BoxFuture<'static, Result<serde_json::Value>>;
+    ) -> BoxFuture<'static, Result<BoxStream<'static, Result<String>>>>;
 
     #[cfg(any(test, feature = "test-support"))]
     fn as_fake(&self) -> &provider::fake::FakeLanguageModel {
@@ -92,10 +92,11 @@ impl dyn LanguageModel {
     ) -> impl 'static + Future<Output = Result<T>> {
         let schema = schemars::schema_for!(T);
         let schema_json = serde_json::to_value(&schema).unwrap();
-        let request = self.use_any_tool(request, T::name(), T::description(), schema_json, cx);
+        let stream = self.use_any_tool(request, T::name(), T::description(), schema_json, cx);
         async move {
-            let response = request.await?;
-            Ok(serde_json::from_value(response)?)
+            let stream = stream.await?;
+            let response = stream.try_collect::<String>().await?;
+            Ok(serde_json::from_str(&response)?)
         }
     }
 }