Add simplecov:disable / simplecov:enable directive comments

[sferik]

Alternative to #1068 implementing the directive shape proposed at #1068 (comment):

# simplecov:disable line
# simplecov:disable method, branch
# simplecov:disable                                        <- shorthand for all three categories
# simplecov:disable line legacy adapter, scheduled to die  <- trailing text is a free-form reason
# simplecov:enable
raise "absurd" # simplecov:disable                         <- inline, single line, no enable needed

Categories are tracked independently. # simplecov:disable line skips those lines from line coverage but still reports any branches and methods that overlap them. # simplecov:disable method flags the method directly without depending on its lines being skipped, which is why Method gains a skipped! setter and an overlaps_with? helper alongside the existing inferred-from-lines skipped?.

Anything trailing the directive is treated as a free-form reason and discarded. The trade-off is that an unrecognised category name silently falls into the reason bucket and the bare form takes effect, so # simplecov:disable cyclomatic ends up over-disabling instead of being rejected. That's a deliberate fail-mode: over-disabling shows up loud in the report (a whole file with no covered lines), whereas silently disabling nothing wouldn't, so typos are easier to catch.

Comment extraction goes through Ripper.lex. That keeps the parser honest about Ruby syntax: a directive marker sitting inside a double-quoted string, a single-quoted string, an interpolated string, or a heredoc body is not a directive, and we no longer pretend otherwise. The # prefix # simplecov:disable line case (whole line is one comment, but the directive isn't at the start of it) is treated as inline, since the directive's column doesn't sit at the start of the comment text.

The new directives subsume what # :nocov: did, so it's deprecated (as is the the configurable SimpleCov.nocov_token). The toggle continues to work but each file that uses it now emits a one-time deprecation warning to stderr at load time pointing the user at the # simplecov:disable / # simplecov:enable replacement. The plan is to remove :nocov: entirely in a future release.

The shape of the implementation:

• SimpleCov::Directive exposes Directive.disabled_ranges(src_lines) which walks the source via Ripper.lex, parses each comment token against a single regex, and returns { line: [Range, ...], branch: [...], method: [...] }. The Directive value object itself is mostly an internal accumulator for disabled_ranges; users of the class don't need to reach for it.
• LinesClassifier#classify consults the line ranges so that SimulateCoverage (used for tracked-but-unloaded files) marks disabled lines as NOT_RELEVANT, matching the behaviour :nocov: already had for that path. Both this file and SourceFile require_relative "directive" directly so that requiring either one in isolation doesn't blow up.
• SourceFile lazily memoises per-category ranges and consumes them in the three process_skipped_* methods. Branches and methods both use overlap-with-range checks, so a directive that sits inside a method body or a branch body still skips the enclosing construct. process_skipped_branches also consults branch.report_line so an inline directive on the condition line of a multi-line if (where the arm body's source range starts on a different line) still skips the arm.
• SourceFile#warn_no_cov_deprecation is the deprecation hook. It dedupes per-filename via a class-level Set (SimpleCov::SourceFile.nocov_warned) so noisy projects only see one line per file rather than one per occurrence.