Introduce Active Learning

[PawelPeczek-Roboflow]

Description

With this PR we introduce foundational changes to enable active learning in inference.
So far, feature is disabled - due to lack of implementation of backend changes at Roboflow platform, but that will change soon.

Features supported:

• sampling strategies (random sampling for now) with parameters configurable via Roboflow app
• limits of collected datapoints
• uploading images with predictions into roboflow app - later those may be assigned as labeling tasks

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• This change requires a documentation update (but let's defer it in time until feature is shipped into Roboflow platform)

How has this change been tested, please provide a testcase or example of how you tested the change?

• running inference server locally (with and w/o Redis) and submitting inferences verifying them with Roboflow APP
• running inference against stream

Any specific deployment considerations

• limits apply to local instance only if global cache in Redis is not connected
• cache keys related to limits will be at max < 10 per project in given moment of time (expiration is set reasonably)
• locking around limits credits updates are within small portions of activities (only to ensure atomic updates) - not around whole registration

Docs

• Docs updated? What were the changes:

[paulguerrie]

This is the place holder waiting for updates on the Roboflow main app back end?

[PawelPeczek-Roboflow]

yes, exactly

[paulguerrie]

Gave it a thorough read, looks really great! Nothing I would change and no glaring bugs I could find. Great work!