Fix digesting of large data (more than MAX_UINT32 bits)

[superquadratic]

This is based on @samKamerer's work in PR #39, with the following additions:

• rebased to the current master
• uses constants instead of calling Math.pow(2, 32) in digest()
• uses exactly 2^32 bits in the test (the smallest failing amount)

Sorry for opening another PR for this, but I didn't know how to do it otherwise.