To make this all work, I had to remove the stringification cache
from matchers. In theory, this can cause performance penalty in
cases where single matcher instance is stringified multiple times,
but in practice this does not happen much, and the difference is
surprisingly small anyway, because the performance of stringification
is already horrible and full of allocating strings just to throw
them away.

The matcher combinators need P2738 from C++26 to be `constexpr`.

Closes #3091

Not being able to filter generators to specific element has been regularly
causing problems. It was possible to use a dynamic section to run tests
for specific element in a generator, at least if the element had a nice
string representation, but the test case would still run once per element
in the generator.

With this change, it is possible to have the generator return only one
specific element, and do so based on the index, rather than the string
representation of the element. This enables simple debugging of tests
that fail for specific generator element.

Specifically, this commit makes the `-c`/`--section` parameter
strictly ordered and hierarchical, unlike how it behaved before,
which was a huge mess -- see #3038 for details.

Closes #3038

As with other Bazel env vars, it overrides the corresponding CLI
parameter if both are set.

Closes #3021
Closes #3024

Adding unreachable to `FAIL` and `SKIP` made Clang become very good
at figuring out that tests of `FAIL` and `SKIP` will never return
and thus could be marked as [[noreturn]].

To avoid introducing lot of warning suppression noise into the
test files, we just disable it instead.

fix #2755

https://cmake.org/cmake/help/v3.16/module/FindPythonInterp.html

2 spaces seems to be the more common indentation level so that's what
I unified around.

This is not necessary so we can remove it. Most conditionals already
omit this anyways.