Print context of violations

[ZedThree]

Example:

warning: test.f90:90:3: P011 prefer 'real(real64)' to 'double precision' (see 'iso_fortran_env')
   |
89 |   ! Should trigger for use of 'double precision'
90 |   double precision function double_prec(x)
   |   ^ P011
91 |     double precision, intent(in) :: x
92 |     double_prec = 2 * x
   |

warning: test.f90:91:5: P011 prefer 'real(real64)' to 'double precision' (see 'iso_fortran_env')
   |
89 |   ! Should trigger for use of 'double precision'
90 |   double precision function double_prec(x)
91 |     double precision, intent(in) :: x
   |     ^ P011
92 |     double_prec = 2 * x
93 |   end function
   |

warning: test.f90:97:1: M001 subroutine not contained within (sub)module or program
   |
96 | ! This function should trigger for missing enclosing module
97 | subroutine triple(x)
   |  ^ M001
98 |   integer, intent(inout) :: x
99 |   x = x * 3
   |

A couple of annoying things:

• Something really weird going on here, where I can't get it to
  put the annotation in the first column: it's either in column 2
  or the end of the previous line. But does appear to be right
  for other columns!
• Some annoyance here: we have to have some level prefix to our
  message. Might be fixed in future version of annotate-snippets
  -- or we use an earlier version with more control.
• Also, we could use .origin(path) to get the filename and
  line:col automatically, but see above about off-by-one error

[LiamPattinson]

Amazing, this looks so much better.

I think I've figured out a fix to the weird off-by-one error, but I haven't tested it enough to be 100% sure.

[LiamPattinson]

Suggested change

	.annotation(Level::Error.span(label_offset..label_offset).label(code)),
	.annotation(Level::Error.span(label_offset..label_offset.saturating_add(1)).label(code)),

[LiamPattinson]

Suggested change

	let label_offset = offset_up_to_line + col.saturating_sub(1).max(1);
	let label_offset = offset_up_to_line + col.saturating_sub(1);

[LiamPattinson]

As an aside, it might be nice to introduce a --concise mode that doesn't include the snippet, as it seems to be taking about 4 times as long when applied to a project I'm using for simple benchmarking (although 2 seconds to produce over 10000 annotated code snippets is still impressively fast as far as I'm concerned).