Make it possible to evaluate package loadability.

[maleadt]

Use that as the fallback for blacklisted packages.

Fixes #271
Will only deploy next week.

[maleadt]

Example report: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/e0656ac/report.html

This will need some additional fixes across the PkgEval stack to ensure status=:ok reason=:loaded isn't rendered as positively as reason=:tested.

@KristofferC Thoughts on this change? It was motivated by having additional coverage on loadability of packages when they're blacklisted because of test failures, but we'll need to be careful that it's clearly shown as such.

[maleadt]

This regressed PkgEval times from 10h to 18h, so I'll likely revert this.

Example run: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/3124e4e_vs_309021d/report.html

[maleadt]

Switching to not using TestEnv, but only compiling the actual package, results in much better timings: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/3236e12_vs_17857ca/report.html

I wonder if somethings up here, or if it's really test dependencies accounting for so much of the total PkgEval time...

EDIT: Oh, looks like GitHub had an outage and many packages weren't tested...

[maleadt]

Here's a more realistic run: https://s3.amazonaws.com/julialang-reports/nanosoldier/pkgeval/by_hash/1dd25f0_vs_5ee67b8/report.html

So it regressed from 10 to 15h, which I guess is fair considering how much additional packages were covered:

In total, 10154 packages were evaluated, out of which 5246 successfully tested, 2723 were not tested but did load successfully, 45 crashed, 1264 failed and 876 were skipped.