fix(autofix): Prevent long wait times for insights threadpool at end …

…of run (#1631)

Issue #1630 

I discovered in profiling that when the root cause step completed, there
was sometimes a huge wait (the case I saw was 6 seconds) for the thread
pool to finish it's last insight card generation, before moving onto the
root cause formatter. (similar for the coding step)

Why? Because we started a new insight sharing thread after every LLM
call, including the final call of the agent run. Due to Zach's changes
that make us wait for threads to complete before ending the run, this
caused big delays.

Solutions in this PR:
1. Don't generate an insight card on the final output of the run. That
was somewhat redundant anyway if we're about to show the final output
card, plus this avoids starting a new thread from scratch that blocks
everything else.
2. Adds a second thread worker. That way while running in parallel with
the main agent, one insight doesn't have to wait for the previous to
complete, and it's unlikely that insights will "fall behind" the main
agent.

We could completely eliminate this delay if we didn't wait for the
threads to be completed, and I'm not sure that's a bad idea, but I'm
still assuming we don't want to sacrifice that safety. This PR mitigates
most of the issue while maintaining thread safety.

src/seer/automation/autofix/autofix_agent.py

@@ -136,9 +136,10 @@ def run_iteration(self, run_config: RunConfig):
         # log thoughts to the user
         cur = self.context.state.get()
         if (
-            completion.message.content
-            and self.config.interactive
+            completion.message.content  # only if we have something to summarize
+            and self.config.interactive  # only in interactive mode
             and not cur.request.options.disable_interactivity
+            and completion.message.tool_calls  # only if the run is in progress
         ):
             text_before_tag = completion.message.content.split("<")[0]
             text = text_before_tag
@@ -171,7 +172,7 @@ def run_iteration(self, run_config: RunConfig):
     @contextlib.contextmanager
     def manage_run(self):
         self.futures = []
-        self.executor = ThreadPoolExecutor(max_workers=1, initializer=copy_modules_initializer())
+        self.executor = ThreadPoolExecutor(max_workers=2, initializer=copy_modules_initializer())
         with super().manage_run(), self.executor:
             yield
         for future in self.futures: