Starting with implementing the Graph traits for the BasicCoverageBlock
graph.

Most of these TODO comments were reminders to remove commented out lines
that were saved but replaced with an alternative implementation.

Now that the new BCB graph is implemented and tests still pass, it
should be safe to remove the older attempts.

I need to change how I determine if a target from a SwitchInt exits the
loop or not, but this almost worked (and did successfully compile and
create the right coverage). It just didn't select the right targets for
the optimized expressions.

Still not fully optimized to remove unneeded expressions, or expressions
that complicate the MIR a bit without really improving coverage or
efficiency.

Getting some coverage results that don't seem right:
  * In three tests, the last line of the function (last brace) is
    counted twice.
  * In simple_match, the { let z; match is counted zero times (and used
    to be uncovered) but how would a developer ever get non-zero coverage
    for this code?

* Avoid adding coverage to unreachable blocks.
* Special case for Goto at the end of the body. Make it non-reportable.

They may still get counters but only if there are dependencies from
other BCBs that have spans, I think.

At least one test, "various_conditions.rs", seems to have simplified
counters as a result of this change, for whatever reason.

I relized the "hack" with GapRegions was completely unnecessary. It is
possible to inject counters (`llvm.instrprof.increment` intrinsic calls
without corresponding code regions in the coverage map. An expression
can still uses these counter values, solving the problem I thought I
needed GapRegions for.

Just moved the file, before refactoring into multiple files in the new
compiler/rustc_mir/src/transform/instrument_coverage/ directory.

Still working on make_bcb_counters() refacoring to split it into
separate functions.

I should have re-tested for these, in the last PR. Here are those
updates.

Compiling with coverage, with the expression optimization, now works on
the json5format crate and its dependencies.

Also removed the SIMPLIFY_EXPRESSIONS option, which added complexity
without tangible benefits.

It could be refactored even further, if that's preferred/requested, but
its much better than it was ;-).