This has caused several issues throughout the years, the lastest
instance being the reopening of #14933, so let's just remove this
feature. The optimization this allows for when rebuilding the same
branch frequently (together with the keeprunning option) was probably
only used by me, and rarely at that, so I'd argue it is not worth the
maintenance cost.

Will-fix: #14933

Since the disk cache shrinking is here to ensure that we don't run out
of disk space, it should actually be done before *each* build.

When pre-building the wiki, we modify the PO files which results in a
conflict from the base branch merge in case it modifies the same
files.

Will-fix: #12611

It seems fuser binary was missing in the build VM for the cleanup code
to correctly work and kill processes that prevent the unmounting of the
build directory.

Will-fix: #12527

Refs: #11972

Otherwise umounting the tmp build dir fails.

Refs: #11972

NTP will never succeed in offline mode, blocking the build forever.

It was a misunderstanding to remove it.

The cached wiki builds only survive until the builder VM shutdown, but
it's still useful for speeding up rebuilds after failures, or
in general with the 'keeprunning' or 'rescue' options.

This is more coherent with the way auto/* and Jenkins name this
variables too.

Refs: #11972

Refs: #11972

We definitely need this enabled on Jenkins. For others, let's not
enable it by default since it has quite a large impact, and might be
surprising.

Refs: #11972

I don't think there's any benefit with `git clone` over this approach.

Except when rescue mode is on.

The old condition, if we were building in RAM, doesn't make sense
since residual disk build directories are equally problematic.

This reverts commit 04bc8bdb.

There's no need for setting this on Jenkins any more.

The reason we added it seems to be wrong, and the operation just takes
longer with `--no-local`.

Apparently we don't need it on Jenkins any more.

... instead of recomputing it with ad hoc re-implementations or things
already present in auto/scripts/utils.sh.

Refs: #11972

Refs: #11972

Otherwise we won't get the refs for the base branches, which we might
want to merge in `auto/build` later.

This is meant to ease debugging.

This branch already assumes it to be the case in some places, and it's
handy since with an actual checkout, we can use scripts, like
auto/scripts/apt-snapshots-serials, that we need. We also adapt the
build-tails script to this change.

(Note: This change is actually already present in another
branch, wip/11972-use-vagrant-in-jenkins branch, but it's not suitable
for merging into this one yet.)

We can avoid a lot of problems by setting http_proxy like we want it
only when explicitly needed since Vagrant itself uses it (which we had
to work around when the in-VM proxy was used).

… in the hope that it tells us why the filesystem cannot be unmounted,
if this ever happens.