Fixes #20029

During the intial NCG development, GHC did not have support for
anything below Words.  As such the NCG didn't support any of this
either.  AArch64-Darwin however needs support for subword, as
arguments in excess of the first eight (8) passed via registers
are passed on the stack, and there in a packed fashion.  Thus
ghc learned about subword sizes.  This than lead us to gain
subword primops, and these subsequently highlighted deficiencies
in the AArch64 NCG.

This patch rectifies the ones I found through via the test-suite.
I do not claim this to be exhaustive.

Fixes: #19993

Metric Increase:
    T10421
    T13035
    T13719
    T14697
    T1969
    T9203
    T9872a
    T9872b
    T9872c
    T9872d
    T9961
    haddock.Cabal
    haddock.base
    parsing001

This adds constructors to TcRnMessage to replace use of
TcRnUnknownMessage in Ghc.Tc.Module.
Adds a test case for the UnsafeDueToPlugin warning.

Closes #19926

Since 58cfcc65 the default for jobs has been "interruptible", this means
that when new commits are pushed to a branch which already has a running
pipeline then the old pipelines for this branch are cancelled.

This includes the master branch, and in particular, new commits merged
to the master branch will cancel the nightly job.

The semantics of pipeline cancelling are actually a bit more complicated
though. The interruptible flag is *per job*, but once a pipeline has run
*any* non-interruptible job, then the whole pipeline is considered
non-interruptible (ref
https://gitlab.com/gitlab-org/gitlab/-/issues/32837). This leads to the
hack in this MR where by default all jobs are `interruptible: True`, but
for pipelines we definitely want to run, there is a dummy job which
happens first, which is `interreuptible: False`. This has the effect of
dirtying the whole pipeline and preventing another push to master from
cancelling it.

For now, this patch solves the immediate problem of making sure nightly
jobs are not cancelled.
In the future, we may want to enable this job also for the master
branch, making that change might mean we need more CI capacity than
currently available.

[skip ci]

Ticket: #19554

Co-authored-by: Matthew Pickering <matthewtpickering@gmail.com>

Cabal explicitly passes options to set the rpath, which we then also try
to set using install_name_tool. Cabal should also pass `-fno-use-rpaths`
to suppress the setting of the rpath from within GHC.

Fixes #20019