Bigint encoding takes quadratic O(n^2) time

[Sekenre]

The code seems to be a long integer time complexity O (n ^ 2)?
Let's say I have a 10,000 bit integer

Originally posted by @fsssosei in #28 (comment)

The repeated divisions and inserting at the beginning of the list get noticeably slow once you get to about 100000 digits.

>>> timeit(lambda: cbor2.dumps(2**200), number=50000)
1.3028403969080813
>>> timeit(lambda: cbor2.dumps(2**2000), number=50000)
12.468400787521446
>>> timeit(lambda: cbor2.dumps(2**20000), number=50000)
710.3362544665841
>>> 

[Sekenre]

I have a fix for this in my repo and will merge it in once I've done some more testing.

The algorithm used for bigint encoding comes from Stackoverflow. and is perfectly reasonable for arbitrary bases in C but carries a lot of overhead in Python.

The most efficient method I've found so far converts the float into it's hexadecimal representation and then unhexlify's the string.

[Sekenre]

Performance improvement on python 3.6

>>> timeit(lambda: cbor2.dumps(2**20000), number=50000)
3.508080048715488

[Sekenre]

Fix has been merged, will be in next release