WIP: use mkdtemp for IPC paths

[jnpkrn]

work in progress:

• core implementation
• code cleanup, doc, valgrind sanitization etc.
• chmod + chown recursion, also reconsider the permissions
  scheme (single perms guard at the dir + lax approach for items,
  vs. everything strict) and sanity checking
  (data contained within the same parent dir header
  enforcement in client)
• rmdir at more places as a proper cleanup
• using gnulib surrogate for mkdtemp where not available natively

[radosroka]

Hi @jnpkrn , is this also going to fix #369 ?

[jnpkrn]

is this also going to fix #369?

For explanation, this was to demonstrate (mainly to @chrissie-c who
originally set off in the ill-devised direction):

#339 (comment)

@radosroka:

You apparently struggle with phenomenon of directory ownership change
(that you seem would expect not to be mangled in contrast how it would
have been by standard process-to-objects permissions transfer), but
rest assure this PR is definitely not going to help (if anything, it
does not handle that problem at all -- while it was apparent it needs
some sort of assessment should it be eventually picked -- see the TODO
notes:

• chmod + chown, also reconsider the permissions scheme

That being said, closing this PR so as not to cast any confusions.