Add a script for diffing cachegrind output

[glbrntt]

Motivation:

When profiling with cachegrind it's often useful to compare changes
across runs. It's reasonably easy to do this for program totals but a
little harder to see changes between instructions.

Modifications:

Add a script for diffing output from cachegrind. Features:

• runs input files through swift demangle (unless you tell it not to)
• sort on file1, file2, or the delta
• ignores changes less than a configurable low watermark (configurable)

Result:

Easier to diff output from cachegrind