Get Lint working on 0.6

[TotalVerb]

The first commit converts DataType to Type everywhere. This should always have been the case, but is especially important in 0.6, as many types like Array are now UnionAlls instead of DataTypes.

The second commit rewrites the version constraint code to not call eval. Not calling eval is a good idea anyway, but is especially important in 0.6 due to the world age restrictions. The calling of new methods created with eval is subject to world age restrictions in 0.6, and removing the eval addresses that issue.

The third commit changes from typeof(x) == T to isa(x, T) in many places. 0.6 introduces new x isa T syntax which would make sense to use in the future, and this is a first step toward that modernization.

The fourth commit addresses the remaining, miscellaneous problems that prevent Lint from working properly on 0.6, and involves some updates to legacy code (this fixes several issues along the way). Overall, Lint should be better at detecting types. Once 0.4 support is dropped, large amounts of code can be removed.

The fifth commit (in progress) fixes breakage of 0.4 and 0.5 introduced by the first four commits.

Fixes #79
Fixes #137

[coveralls]

Coverage decreased (-1.8%) to 86.272% when pulling 485bf81 on TotalVerb:fw/fix-0.6 into 9028f4f on tonyhffong:master.

[coveralls]

Coverage decreased (-1.8%) to 86.272% when pulling d38b027 on TotalVerb:fw/fix-0.6 into 9028f4f on tonyhffong:master.

[coveralls]

Coverage decreased (-1.8%) to 86.272% when pulling d05da1b on TotalVerb:fw/fix-0.6 into 9028f4f on tonyhffong:master.

[TotalVerb]

There was some code incorrectly supposing that slicedim always returns either a vector or a matrix, when in fact this depends on the values of the arguments. Removing that code (and test) is necessary for the tests to pass on 0.6, where inference is used to guess the result type instead.

[coveralls]

Coverage decreased (-1.2%) to 86.024% when pulling d7ac0cd on TotalVerb:fw/fix-0.6 into b0005ca on tonyhffong:master.

[coveralls]

Coverage decreased (-1.2%) to 86.024% when pulling d7ac0cd on TotalVerb:fw/fix-0.6 into b0005ca on tonyhffong:master.

[TotalVerb]

All the tests are passing and I think this is ready to go after review!

@tonyhffong
@Michael-Klassen

Once this is in I'll have another PR fixing up minor issues and cleaning up some loose ends.

[Michael-Klassen]

Keep the wording on this error message or change it to be more descriptive.

[TotalVerb]

The error message is actually the same. I had to change the test because it prints as Type{T} on 0.5 or Type on 0.6, and I didn't want to select which based off version.

[TotalVerb]

I think, going forward, we will want to standardize the error messages based off error codes, and then disassociate the testing of linting creating the right error code, and testing the error codes being associated with an appropriately descriptive message.