fix mxbai reranker crash in chat TUI

CHANGELOG.md

@@ -5,6 +5,10 @@
 
 - Documentation generator swapped from `mkdocs-material` to `zensical`. Drops `mkdocs` / `mkdocs-material` dev deps, replaces `mkdocs.yml` with `zensical.toml`, adds `overrides/main.html` (OG/Twitter share meta) and `docs/stylesheets/extra.css`. `build-docs` workflow now runs `uv run zensical build` and publishes via the GitHub Pages artifact actions instead of `mkdocs gh-deploy`.
 
+### Fixed
+
+- mxbai reranker crashing inside the chat TUI with `ValueError: bad value(s) in fds_to_keep`. tqdm constructs a `multiprocessing.RLock` on first use, whose `resource_tracker` spawn picks up `sys.stderr.fileno()`; Textual's redirected stderr returns `-1`, failing the `fork_exec` validation. The reranker now pins tqdm's class lock to a `threading.RLock`.
+
 ## [0.48.0] - 2026-05-20
 
 ### Added

haiku_rag_slim/haiku/rag/reranking/mxbai.py

@@ -1,11 +1,20 @@
 import asyncio
+import threading
 
+import tqdm
 from mxbai_rerank import MxbaiRerankV2  # pyright: ignore[reportMissingImports]
 
 from haiku.rag.config import Config
 from haiku.rag.reranking.base import RerankerBase
 from haiku.rag.store.models.chunk import Chunk
 
+# tqdm's default class lock is a multiprocessing.RLock; constructing it spawns
+# resource_tracker, which inherits sys.stderr's fileno. Inside Textual's chat
+# TUI, sys.stderr.fileno() returns -1, landing in fds_to_keep and failing the
+# fork_exec validation. A threading lock is sufficient since we never share
+# tqdm progress bars across processes.
+tqdm.tqdm.set_lock(threading.RLock())
+
 
 class MxBAIReranker(RerankerBase):
     def __init__(self):