Optimize generated subtargets' interaction with SpecifiedSourceFiles

[Eric-Arellano]

Benchmark runs ./pants --no-pantsd lint 'src/python/pants/util/*.py' 10 times without pantsd and with Process already cached.

Before:

            Mean        Std.Dev.    Min         Median      Max
real        2.697       0.034       2.652       2.706       2.752
user        2.354       0.025       2.317       2.361       2.392
sys         0.432       0.011       0.420       0.428       0.452

After:

            Mean        Std.Dev.    Min         Median      Max
real        2.605       0.018       2.588       2.599       2.648
user        2.259       0.015       2.243       2.253       2.296
sys         0.403       0.005       0.395       0.405       0.408

This means that we almost never use SpecifiedSourceFiles's logic. Almost always, SpecifiedSourceFiles is the same thing as AllSourceFiles. The only time that the two diverge is when file arguments cannot use generated subtargets because there are multiple owners of the file.

[ci skip-rust-tests]

[coveralls]

Coverage remained the same at 0.0% when pulling b0021a1 on Eric-Arellano:specified-snapshots into 0a61cf8 on pantsbuild:master.