API/DOC/backwards: rework config defaults

[stsievert]

What does this PR implement?
It changes the defaults for Salmon.

• API: the values change, and there's also some easier methods to pass arguments to every sampler.
  • API: allow customizing the HTML title of query page
  • API: renames alg_ident to sampler.
• DOC: the documentation around the defaults is much better.
• DOC MAINT: including...
  • also adds to "why adaptive?"
  • cleans up docs around init file structure
  • clarifies samplers with same name.
  • expands upon examples (shows/tests default config, shows customization, etc)
• TST: the config are tested too!
• TST BUG: fixes CI testing! (required some backend work to make sure old endpoints weren't being hit after being deleted).
  • MAINT: properly sets number of processors
• API: now YAML in /config can be uploaded to /init.
  • (technically some backwards incompatibility; n has been removed from /config).
• TST: makes sure configs are proper experiments.
• MAINT: only change docs on release (lesson learned)
• MAINT/DOC: cleans up Runner/Sampler

Reference issues/PRs
Closes #124, closes #82 and closes #115.

[stsievert]

Some recommendations:

• Make targets/ZIP file more clear (larger, earlier, etc).
• Add clarification around the names in testing: ARR: {}.
  • mention this more explicitly in "Sampler configuration"?
• Add link to "sampler configuration" (https://docs.stsievert.com/salmon/samplers.html) in API/Config and on "getting started" page.
• Make more clear what "sampler" and "sampling" control.

[stsievert]

All tests pass locally for commit 72152a2. On CI, one test is failing (looks like a left-over from a previous test to make sure backend errors are logged).

[stsievert]

I've fixed the CI tests, and have accelerated them.

• Fixing CI: launch max(4, NUM_PROCS) instead of NUM_PROCS Dask workers.
  • This is a problem because the GitHub Actions Docker machine only had 2 processes, meaning I'd get an index error when trying to access workers[2].
• Acceleration: reduce sleeps. Biggest change: not sleeping while running rj.reset when hitting /reset.
  • This change went from 665s to 130s on my local machine

The time spent in the CI went from 37-38 minutes to 14-15 minutes.

[stsievert]

I've reworked resetting samplers. I think this closes #82. After hitting /reset, samplers would still live inside the FastAPI app. I reworked that by using a global variable on the condition that the backend server is run with one worker (and I added a note to the algorithm dev docs).

As a result, the CI works now.

[stsievert]

Now, hitting /reset?force=1 completely resets Salmon. I've modified the docs to reflect that, and have also reset the username/password (and have reopened #115).