Update version bounds for GHC 7.8

[bennofs]

I bounced the version bound for base to < 5, as it will cause issues with every new
ghc version and the package doesn't use any exotic functions from base. I removed the
upper bound on template-haskell, because it never makes sense to install a different
version of template haskell than the one provided by GHC.

[hvr]

Tbh, I'd like to phase out deepseq-th in favor of deepseq-generics which I regard a much better approach to the problem it tries to solve. I regard this package obsolete.

Is there something which you can only do with deepseq-th?

[bennofs]

Not really, I could just switch to deepseq-generics.

[hvr]

ok; fwiw, deepseq-generics also is more reliable

[maxcan]

Making the switch over now. But, doesn't deepseq-generics impose a runtime penalty that deepseq-th would impose at compile time?

[hvr]

@maxcan that's the cool thing about deepseq-generics: The GHC optimizer is clever enough so that the resulting code is as good as the one resulting by TH (if not even better in some cases I didn't handle in the TH code).

If you happen to find cases where this is not the case, it's a bug in the GHC optimizer. So you won't be able to measure a runtime penalty, but if you do, please file a bug, as that's most likely a bug in GHC then.