Rewrite the sage-cleaner

[vbraun]

Its still bad. Here is a wishlist:

• don't continuously restart the sage cleaner during doctests
• make it part of the sage library to get rid of the code duplication for temporary directories that keeps breaking the cleaner
• in particular, make it doctestable
• Entries in spawned_processes are killed by the cleaner and by sage.interfaces.quit.expect_quitall().
• avoid potential races with PID wrap-around: Sage should announce its shutdown to the cleaner.
• Ensure (and doctest) that the temp directory with given PID is deleted before Sage starts. Even if the cleaner doesn't work we must not kill stuff from old spawned_process files.

[leif]

Replying to vbraun:

• make it part of the sage library to get rid of the code duplication for temporary directories that keeps breaking the cleaner

Not sure if that makes sense. People should take more care when changing such parts... ;-)

In any case, "both parts" should of course somehow use the same interface (i.e., directories etc.), which doesn't or shouldn't require code duplication. It should IMHO still be possible to run a single instance of the Sage cleaner for different (running) Sage versions, or, at the very least, the cleaner(s) shouldn't break when running different Sage versions at the same time.

[jhpalmieri]

My suggestion on #14055 was to put, for example,

from sage.env import HOSTNAME

in sage-cleaner.