internal/server/singleprocess/state: Add ability to prune jobs and deployments

[evanphx]

This limits deployments and jobs to 10,000 records each. The code only prunes the data from the memory indexes, it leaves in the disk.

The pruning of data is done on a regular interval rather than inline during data inserts (ie, it's like a typical garbage collection process). The pruning logic is shared between app records and jobs as well.

Pruning the data from disk would come later.

[mitchellh]

• I like only pruning the indexes. That way if we mess anything up we aren't deleting real data.
• I'd introduce total record counts as package vars to avoid the iteration in jobs since that seems pretty easy.
• For the app ops, I feel like we should just use the cursor at all times rather than ForEach. The existence of the secondary unmarshal, index, seq number tracking, etc. logic makes me nervous. Seems easy to introduce a bug that only exists in one or the other "mode".
• Similar to above with jobs, I'd add functions to *Job that is like IsComplete() bool rather than reproducing that state logic over and over where its easy to introduce a bug.

[evanphx]

1. No prob adding total record counts 👍
2. I'll dry up that iteration code 👍
3. Totes, another good drying 👍

[mitchellh]

This did end up really nice. Great job. A few non-trivial comments but this is great stuff.

[mitchellh]

If you rebase this on main the tests should be fixed 👍