Mitigate PageIndex file-descriptor buildup during long PDF indexing

[plasma16]

Purpose

Reduce risk of [Errno 24] Too many open files during long-PDF indexing with retries.

Changes

• Add best-effort PageIndex client cleanup helper in openkb/indexer.py.
• Create a fresh PageIndexClient per retry attempt.
• Explicitly close backend storage after failed attempts and at function exit.
• Trigger gc.collect() between failed attempts to accelerate descriptor release in long runs.

Why this helps

In local mode, repeated failed/retried indexing can leave resources alive longer than expected. This patch ensures resources are released promptly per attempt.

Test evidence

• python3 -m py_compile openkb/indexer.py (pass)
• Full pytest suite not run in this environment because pytest is not installed in system Python.

Notes

This is an OpenKB-side mitigation while upstream pageindex resource handling can be further hardened.