A case-insensitive MatchCI module without dependencies.

[mikehat]

Here's a version of MatchCI without the case-insensitive dependency. I kind
of dislike the HasCase and caselessEq names, but didn't take time to think
of any that didn't overlap some other lib's. Keeping them internal (aka not the
first commit on the match-ci branch) helps.

There are actually three commits on this branch. The first adds a class to the
StringLike module, which may not be desirable. The second moves all changes
into the MatchCI module. The third approach adds a new module Text.HasCase
with the option of not making it an exposed module.

Here's why I think MatchCI is better than a toLower :: Tag s -> Tag s:

• Not all uses of Match are necessarily against static names that the client
  code controls. Assuming that the client code can match a toLower-ed tag
  stream against lower-case literal values assumes that the client code
  contains the names to be matched. Of course, the client can toLower
  everything, which leads to...
• Client code that doesn't match against string literals will be littered with
  toLower. Switching to case-insensitive matching becomes as simple as
  importing the MatchCI module rather than Match.
• Using toLower to make everything work means lower performance. We can be
  sure that the penalty is small, but performance doesn't end at tag parsing.
  It is reasonable to assume that almost every opening tag will be matched for
  a tag name. While the Match module isn't part of the bench routine, any use
  against real-world HTML will require case-insensitive matching. I think that
  for this usage (compare once and forget), a local function is probably better
  than using the case-insensitive package (allocate a lower-case instance or
  a thunk).

[ndmitchell]

Thanks for the revised pull request. You arguments in favour of MatchCI rather than my suggestion seem good.

Final remark is why not merge HasCase into StringLike, and demand everyone providing StringLike implements caselessEq? I don't see that as particularly burdensome (there are probably no implementations of StringLike outside tagsoup), and simplifies the number of pieces.

[mikehat]

I originally had caselessEq inside StringLike, but didn't want to break anyone's instance. But I think you're right. How about a better name, ciEq or ^==?

[ndmitchell]

In fact, you can have a default instance for caselessEq which does toString and caselessEq that way - then no one who doesn't implement it gets broken.

I'm not too fussed about the name, but given it's pretty rare, I would avoid symbols or excessive/unusual abbreviations. caseInsensitiveEq is probably where I'd land, but it's up to you.