ENH: Improve possible dataset error discoverability

[larsoner]

Before merging …

• Changelog has been updated (docs/source/dev.md.inc)

Our default of mne_log_level = "error" I think is unsafe -- it's too easy for users to hit likely errors in their datasets and not notice. So this PR:

1. Changes the default to "warning"
2. Updates some option values that should be usable
3. Adds a ignore_warnings config param to make it easy to choose warning messages to suppress
4. Fixes us actually setting our own internal logging level.
5. Moves read_raw_bids_verbose, mne_log_level, and log_level into exec_params, which is the correct scope for them: they should not affect run caching (previously at least read_raw_bids_verbose would have) and outputs/results, just the messages that get printed to the console

I think evidence for the utility is already here in this PR when combined with our tests treating warnings-as-errors:

1. See the ignore_warnings I ended up adding for ds0001117 -- you can see multiple issues with it as a BIDS dataset and/or our reading of it
2. See the change to _07_make_epochs.py, where I needed a new _ignore_warnings context (which caused me to have to think about whether or not it was appropriate/safe!)

I suspect as all datasets run through I'll hit a lot more warnings for them and need to add ignore_warnings to a number of them. But we'll see...

[larsoner]

Okay I think this is ready @drammock !