This implements support for applying PGO to the rustc compilation step (not
standard library or any tooling, including rustdoc). Expanding PGO to more tools
is not terribly difficult but will involve more work and greater CI time
commitment.

For the same reason of avoiding greater time commitment, this currently avoids
implementing for platforms outside of x86_64-unknown-linux-gnu, though in
practice it should be quite simple to extend over time to more platforms. The
initial implementation is intentionally minimal here to avoid too much work
investment before we start seeing wins for a subset of Rust users.

The choice of workloads to profile here is somewhat arbitrary, but the general
rationale was to aim for a small set that largely avoided time regressions on
perf.rust-lang.org's full suite of crates. The set chosen is libcore, cargo (and
its dependencies), and a few ad-hoc stress tests from perf.rlo. The stress tests
are arguably the most controversial, but they benefit those cases (avoiding
regressions) and do not really remove wins from other benchmarks.

The primary next step after this PR lands is to implement support for PGO in
LLVM. It is unclear whether we can afford a full LLVM rebuild in CI, though, so
the approach taken there may need to be more staggered. rustc-only PGO seems
well affordable on linux at least, giving us up to 20% wall time wins on some
crates for 15 minutes of extra CI time (1 hour up from 45 minutes).

The PGO data is uploaded to allow others to reuse it if attempting to reproduce
the CI build or potentially, in the future, on other platforms where an
off-by-one strategy is used for dist builds at minimal performance cost.

This commit switches the x86_64-gnu-nopt builder to use Ubuntu 20.04,
which contains a more recent gdb version than Ubuntu 16.04 (newer gdb
versions fix a bug that Split DWARF can trigger, see
rust-lang/rust#77177 for motivation). x86_64-gnu-nopt is chosen because
it runs compare modes, which is how Split DWARF testing is implemented
in rust-lang/rust#77177.
Signed-off-by: NDavid Wood <david@davidtw.co>

add enable-full-tools to freebsd builds to prevent occasional link errors when compiling rust programs

This was landed on master instead of beta!

It seems that by default the 32-bit headers are not actually installed
when installing development tooling. As we're using gcc headers,
we need to install them as an extra package.

See for reference:
- https://stackoverflow.com/a/54082790
- https://askubuntu.com/a/106092

Also removed the now unused arm tooling

Now with LLVM 9 being the minimum supported version, we can
finally remove the hacks in the dockerfile.

This wasn't in the main PR bumping the version as I didn't quite
understand what's going on and needed here.

apparently llvm-8-tools already had llvm-8-dev as a dependency
which was removed in llvm-9-tools, so we need to explicitly pull
llvm-9-dev to make a build

This bumps the minimal tested llvm version to 9.
This should enable supporting newer LLVM features (and CPU extensions).

This is helpful to catch slightly more bugs before things hit main CI, and
doesn't cost too much extra CI time.

Historically we've disabled these assertions on a number of platforms with the
goal of speeding up CI. Now, though, having migrated to GitHub actions, CI is
already pretty fast, and these debug assertions do bring us some value.

This does leave in some debug assertions that are performance-related: macOS
currently hovers at just under 2 hours.

There are also some other builders which have debug and LLVM assertions
disabled:

llvm-8, PR builder:

In one view, this builder tests our support for older LLVMs. But in reality, a
lot of our tests already disable themselves on older LLVMs, and I think our
general stance is that we really only support the in-tree LLVM. Plus, we really
want CI times on this builder to be really low, as it's run on *every* PR --
that's a lot of CI time.

test-various:

This disables debug asserts still -- as noted in the Dockerfile, we test code
size, and we need debug asserts off for that to work well.

This was recommended by GitHub Support to try reducing the things that
could've caused #78743. I checked the changelog and there should be no
practical impact for us (we already set an explicit fetch-depth).

While technically the i686-unknown-freebsd target has been a tier 2
development platform for a long time, with full toolchain tarballs
available on static.rust-lang.org, due to a bug in the manifest
generation the target was never available for download through rustup.

The infrastructure team privately inquired the FreeBSD package
maintainers, and they weren't relying on those tarballs either, so it's
a fair assumption to say practically nobody is using those tarballs.

This PR then removes the CI builder that produces full tarballs for the
target, and moves the compilation of rust-std for the target in
dist-various-2.

The x86_64-unknown-freebsd target is *not* affected.

The former `ubuntu:19.10` reached EOL in July, 2020, whereas
`ubuntu:20.04` is an LTS release supported until 2025.

These are non-dist CI images, so the impact should be low.

This bumps the version of the bbl bootloader not to perform 64-bit
accesses to the PLIC. Doing so resulted in the QEMU test machine to fail
to boot:

bbl loader
../machine/mtrap.c:21: machine mode: unhandlable trap 7 @ 0x0000000080001f6e
Power off
Signed-off-by: NTom Eccles <tom.eccles@codethink.co.uk>

See GitHub's blog post on why the change was necessary:

    https://github.blog/changelog/2020-10-01-github-actions-deprecating-set-env-and-add-path-commands/

Trying to fix a problem in CI. Maybe some version of Docker is not
passing '' args correctly?

This allows us to make use of a YAML anchor when specifying the try
builder config.