feat: Find out if data generation jobs have started sooner

[justinvdm]

UX changes

• The progress shown for Snaplet AI is now the combined progress percentage for both data generation and shape generation (previously the percentage only counted for data generation)
• When skipping waiting the Snaplet AI steps, in addition to being able to fetch the data generation work that has already completed (which was already previously possible), they will now also be able to fetch the shape prediction work that has already completed

Code changes

• Use startPredictionJobs route for starting up all jobs (both shapes and data generation), removes code that was using startDataGenerationJobs instead
• Removes the brittle hack where we wait for jobs to start before waiting for them to complete, now that we know from the start that the jobs would exist already when we poll for waiting for their completion
• Provides description, initial examples and sample size (from fingerprint config, if specified) to startPredictionJobs route, so we can use this endpoint for the use case of prompt details provided via seed config
• Some related code cleanup that was now possible after these ^ changes

Corresponding snaplet PR: https://github.com/snaplet/snaplet/pull/2638

Demo: https://www.loom.com/share/2665a08668a74eabb3b1680a3b9830fd?from_recorder=1&focus_title=1

[linear[bot]]

S-2112 Find out if data generation job has started sooner

At the moment, we use some guesses to determine whether we need to wait for data generation jobs to start, and then poll until they have. This guesswork isn't robust though, and there are edge cases where it won't be the case. In these cases, the users will either not get the AI results they need, or wait for them indefinitely. Fixing this will also enable us to show progress for both shape prediction and data generation together (e.g. as a single percent), which means the user isn't sometimes seeing progress and sometimes not.