fix(watcher): reclaim mimalloc pages after in-process fallback reindex

watcher_index_fn routes re-indexing through the supervised subprocess (#832)
so the long-lived server hands RSS back to the OS on each cycle. But the
in-process degrade fallback — taken when the supervisor is off or the child
spawn fails — ran cbm_pipeline_run()+cbm_pipeline_free() without the paired
cbm_mem_collect() that every other post-pipeline-run site performs (mcp.c:5758,
pipeline.c:834, pass_parallel.c:799/995, graph_buffer.c:1678).

Because this fallback is the watcher's unattended path and fires on every poll
while the working tree is dirty, the worker-thread pages mimalloc abandons at
pool teardown were never returned. RSS ratcheted into tens of GB overnight
indexing a small repo (observed: ~55GB private commit on ~200 files, Windows,
where the supervised spawn degrades to in-process). Add the collect() to match
the established pattern and bound the fallback path's peak.

Signed-off-by: Greg Illingworth <g.illingworth86@gmail.com>

Author: gillingworth86

src/main.c

@@ -198,6 +198,12 @@ static int watcher_index_fn(const char *project_name, const char *root_path, voi
 
     int rc = cbm_pipeline_run(p);
     cbm_pipeline_free(p);
+    cbm_mem_collect(); /* #832: return mimalloc pages to OS after the in-process
+                        * fallback re-index — mirrors the free()+collect() pairing at
+                        * mcp.c:5758/pipeline.c:834. This is the watcher's unattended
+                        * degrade path (supervisor off or spawn failed); without the
+                        * collect, worker-thread pages mimalloc abandons at pool
+                        * teardown are never returned, so RSS ratchets every poll. */
     cbm_pipeline_unlock();
     return rc;
 }