Improvements around realistic crowdsourcing simulations

[stsievert]

What does this PR implement?
Over in salmon-experiments, I plan to run a set of experiments that use realistic timing for simulations (thanks to the data I have collected). This PR revolves around those improvements. The changes include:

• MAINT: Reworking examples
• MAINT: remove duplicate queries (queries [0, 4, 5] and [0, 5, 4] are not unique because the last two objects get switched randomly).
• MAINT: clean dashboard
  • only show endpoint timings if more than one hit
  • reorganization
  • reduce size of dashboards (don't plot every model update; plot the median over 2 minutes)
• ENH: show histograms of the response {rate, gap} on the dashboard.
• MAINT: don't run RR or RandomSampling (but set the stopped variable in the DB right away)
• MAINT: launch number of workers equal to number of cores (with n_threads=1).
• DOC: clean docs for Sampler, SOE
• API: set default mu in CKL to be 0.05 (as per the original NEXT paper)
• MAINT: run isort, black
• API: change salmon.triplets.offline.OfflineEmbedding's default noise_model to CKL.
• (private) API: allow setting n_search in AdaptiveRunner.
• MAINT: figure out deps (remove numba, etc)

TODO:

• check to make sure n_cores workers are launched.

Reference issues/PRs

[stsievert]

In total, the time required for the CI run to complete has gone from 22–23min to 15–16min, largely thanks to these basic speed enhancements:

• Setting stopped-random right away in e32947c: tests in salmon/tests: 7.35min → 4.92min.
• Basic cleanup around deps in salmon.yml in 6bab90f:
  • installing deps: 5.53min → 3.53min.
  • Building the Salmon server w/ Docker: 8.6min → 6.38min.

[stsievert]

This PR is almost ready for merge. I've update the docs to represent some experiments in salmon-experiments. In short, I show a clear benefit to Salmon's active sampling:

Some jobs to finish; all the active runs should go to 20,000 responses.

[stsievert]

Salmon is working on my own machine; I think the tests have a bug (sigh).