add caching

[atsyplenkov]

Smth like follows but needs more thorough plan:

• for every processed pdf compute hash
• use hashes to store raw GLM-OCR output in a sql database or JSONL
• if the paperdown was asked to process the file with the existing hash, reuse the cached output instead of parsing file one more time

This will require additional --force flag to bypass the caching and new set of commands:

paperdown cache
paperdown cache rm
...

See scoop cache commands or brew cache