Relax some restrictions on validators

[CreepySkeleton]

Also relaxes 'static restriction on validators.

Closes #1771

This PR also relaxes the 'static restriction on the validator fn, thus allowing user to use proper closures that capture some context as validators. This should address (and potentially close) #572.

[CreepySkeleton]

This PR has grown beyond the simple Send + Sync. New changes:

• I have made it possible for validators to capture and mutate context. Since we need it to be working along with Send + Sync, validators form now on use atomic synchronization (single mutex acquire-release pair per validator call). I don't think this makes them any slower since acquiring-releasing a mutex is cheap if there's no real competition, and we only ever use one thread.
• I replaced all the incomprehensible 'b, 'c, 'z... lifetimes with meaningful names. It's really easy to get confused by the lifetimes and use wrong ones when more than one is involved - we have had a few examples of that in the form of &'b Arg<'b> which is wrong.

[pksunkara]

@CreepySkeleton Can you please rebase this?

[CreepySkeleton]

I think I'll extract lifetime renaming toa separate PR and then rebase this one onto that.

[CreepySkeleton]

ping @pksunkara

Pfff, what can I say? Lifetimes in tandem with impl Iterator return types doesn't seem to work very well in some circumstances, see rust-lang/rust#34511 (comment). At any rate, I have rearranged debug assertions a bit. I think the code is more organized and understandable this way.

I have also found a number of wrong tests thanks to the new assert structure (they cover more cases now) which is a point in favor of the rewrite.

Otherwise, I think this is ready and can be merged.

[pksunkara]

bors r+

[bors]

Merge conflict.

[CreepySkeleton]

bors r=pksunkara

[bors]

Build succeeded:

• CI
• clap-rs.clap
• continuous-integration/travis-ci/push