`llm replicate fetch-predictions` command

[simonw]

Refs:

• 11

TODO:

• [x] Progress bar
• [x] Ability to run without fetching everything again
• [x] Tests
• [x] Documentation

[simonw]

Notes on ability to run twice without fetching everything here:

• https://github.com/simonw/llm-replicate/issues/11#issuecomment-1644167678

For the first shipped version I'm going to always fetch every predictions paginated page, but then individual predictions will only be fetched if they're not in the DB yet or are incomplete.

Incomplete means completed_at is null and the status is something other than failed.

[simonw]

[simonw]

Documentation: https://github.com/simonw/llm-replicate/blob/840ea3a0b6182bf9b4ca24a20d0cc0aed072e07a/README.md#fetching-all-replicate-predictions