replace parseFloat with native implementations

[jvasileff]

parseFloat() currently has some problems with precision and errors:

• crash on valid Float values ceylon.ast#98
• ceylon-sdk / source / ceylon / json: Is Numeric overflow desired behaviour? eclipse-archived/ceylon-sdk#296
• Float.integer: show float value in OverflowException #756
• confusing error message for parseFloat() failures #785

Given how much effort would be required to write a really good parseFloat(), especially one that is cross-platform, I suggest (at least for now) we replace the current implementation with native implementations that delegate to platform apis.

[gavinking]

Well I think I've resolved the issues with precision. Now, surely my implementation in Ceylon is significantly slower than the native implementations would be, but we have not actually tested anything to see how much difference there really is.

[jvasileff]

It looks like the new impl solves the critical problems, and the only use case for improved precision I can think of would be minimizing accumulated error for float->string->float roundtrips. But the lack of precision does remain, and I'd hazard to guess that on JS, the difference in performance may be 10x, 100x, or more.