[TD] Script to determine which reverts are caused by bad TD (#6911)

Honestly a pretty messy script

Determines a commit was caused by bad TD, basically:
1. find revert
2. find originally merged commit
3. find last commit on the PR before the merge
4. check merged commit for failures
5. check if failure was excluded by TD on the last commit on the PR

Obviously not perfect, but spot checking it does seem to be ok

Counts are pretty volatile if granularity is too small (week granularity
is very volatile), so I'm not sure if this can really be displayed.
Maybe should make this periodic and still upload to clickhouse? idk

End of the output looks like this but theres a lot of extra output
before it to help debug
```
CAUSED BY BAD TD: 27 / 184 = 14.67%
Unable to check (lack run id) on PR: 1 / 184 = 0.54%
Total caused by bad TD: 27 / 184 = 14.67%
Month 674: 13 bad TD / 77 total = 16.88%
Month 675: 14 bad TD / 107 total = 13.08%
```

Also make clickhouse.py client able to be used in thread pool executor
(http client didn't like having 1 http client for multiple threads i
think)

Author: clee2000

tools/torchci/clickhouse.py

@@ -1,14 +1,12 @@
 import json
 import os
-from functools import lru_cache
 from typing import Any, Dict, List, Optional
 
 import clickhouse_connect
 from clickhouse_connect.driver import Client
 from torchci.utils import cache_json, REPO_ROOT
 
 
-@lru_cache(maxsize=1)
 def get_clickhouse_client() -> Client:
     endpoint = os.environ["CLICKHOUSE_ENDPOINT"]
     # I cannot figure out why these values aren't being handled automatically