Make AtExit a stack and make DirectoryTemporary more robust

[ChrisJefferson]

This is kind of 2 PRs merged together, can split if required.

Firstly, the error message when cleaning up a DirectoryTemporary if it was already deleted wasn't useful, and other temporary directories wouldn't be deleted after the first failing one. This could easily occur if a temporary directory was made, then IO_fork was called.

Secondly, while debugging this I was having some problems which turned out to be caused by the fact the AtExit functions are run in-order. This is against how exit handlers normally work, as it means the first handlers GAP installs (which do things like flush buffers) get run before later, or user-installed handlers.

I pop the list of AtExit handlers, rather than just calling Reverse on the list, in case new handlers are installed as GAP is closing.

[fingolfin]

Hehe, nice comment :)

[codecov]

Codecov Report

Merging #2856 into master will decrease coverage by <.01%.
The diff coverage is 85.71%.

@@            Coverage Diff             @@
##           master    #2856      +/-   ##
==========================================
- Coverage    76.1%   76.09%   -0.01%     
==========================================
  Files         480      480              
  Lines      240949   240955       +6     
==========================================
- Hits       183363   183362       -1     
- Misses      57586    57593       +7

Impacted Files	Coverage Δ
lib/session.g	59.25% <100%> (+0.76%)	⬆️
lib/files.gd	63.63% <75%> (+0.09%)	⬆️
hpcgap/lib/hpc/stdtasks.g	63.42% <0%> (-1.54%)	⬇️
src/iostream.c	62.35% <0%> (-1.15%)	⬇️
src/gap.c	84.2% <0%> (-0.15%)	⬇️
lib/package.gi	71.17% <0%> (+0.02%)	⬆️
src/listfunc.c	94.31% <0%> (+0.17%)	⬆️
src/streams.c	77.64% <0%> (+0.39%)	⬆️