Add lexer for LLVM's MIR format

[dsandersllvm]

MIR is a human readable serialization format that's used to represent LLVM's
machine specific intermediate representation. It allows LLVM's developers to
see the state of the compilation process at various points, as well as test
individual pieces of the compiler. Our documentation for the format can be
found at https://llvm.org/docs/MIRLangRef.html.

Adding a lexer for this format will allow the LLVM documentation to contain
syntax highlighted examples of LLVM-MIR. Two lexers are included in this
change. 'llvm-mir' lexes the overall document format and delegates to 'llvm' and
'llvm-mir-body' as appropriate. 'llvm-mir-body' lexes the contents of the 'body:'
attribute and can be used directly to syntax highlight code examples without
including the document boilerplate.

Since the 'llvm-mir' lexer delegates to the 'llvm' lexer at times, this change
also adds the 'immarg' and 'willreturn' keywords to the 'llvm' lexer as these
were missing.

[Anteru]

This also needs a small example file so it gets tested automatically, a single function in MIR will do it. Please add that to pygments/tests/examplefiles/. Otherwise, this looks good!

[dsandersllvm]

This also needs a small example file so it gets tested automatically, a single function in MIR will do it. Please add that to pygments/tests/examplefiles/. Otherwise, this looks good!

I've created an artificial example as taking a real file from LLVM would have licensing issues to figure out. This example conforms to the format but isn't exactly a valid file as it breaks some constraints that LLVM would normally enforce and isn't for any particular LLVM target. I think it will suffice as a test though.

[dsandersllvm]

Hmm, the test environment seems to be a bit fussier than pygmentize. I've reproduced those test failures locally and I'll investigate

[dsandersllvm]

Hmm, the test environment seems to be a bit fussier than pygmentize. I've reproduced those test failures locally and I'll investigate

The error token was easy enough to fix by adding:
(r'[-+]', Operator),
to the 'mir' state. However, it then failed the round trip test. It turned out I'd managed to use bygroups(using(...)) in a way that resulted in the same newline character generating multiple tokens. The round trip test would then fail on the doubled newlines. I've modified both patterns that use bygroups(using(...)) to fix this

[Anteru]

Thanks for adding the tests -- and the fixes. I'll give it a spin and will merge it then.

[dsandersllvm]

Thanks

[Anteru]

Merged, thanks for the contribution!