Transparent arbitrary-precision integers in the runtime

[lukaszcz]

Three representations of integers with transparent automatic conversion between them.

• Unboxed 31 bit ints.
• Boxed 64 bit ints.
• GMP-based arbitrary-precision integers.
  • Implemented with slightly modified mpz part of mini-gmp added to the runtime sources.
  • Requires malloc-like allocator in the runtime #1557 for GMP memory management.
  • Give the user an option to link with the full GMP library (much better performance: up to 10x faster for numbers with up to a few hundred bits, asymptotically faster for larger numbers).

[lukaszcz]

We can't actually use mini-GMP because it's on LGPL and we don't want that in the runtime. We could use bigz instead (it's on simplified BSD): https://github.com/starseeker/bigz/tree/master