Add mutation testing to CI

[tomato42]

related to #131

0.14.1: With just 20 mutants executed in 25 minutes, it's not really useful (the confidence interval is way too large)
0.15-beta (8deb089): Single test suite execution takes about 9.97s on Travis, in 25 minutes we execute 107 tests, so better, but that still translates to a 99.9% confidence interval of 5.39% to 26.38% for survival rate. So we will need a faster test suite still.
0.16.0: No significant change, about 152 tests in 25 minutes,

todo:

• implement @slow decorator (pytest docs)
• change hypothesis settings when option to run slow tests is enabled to execute fewer examples
• mark all the slow tests, including hypothesis tests, that take more than few miliseconds to execute, with the @slow decorator
• create new static test cases so that branch coverage doesn't fall with the slow tests disabled
• implement small curve (Add support for 112, 128, and 160 curves #24, Add support for small curves #223), use it, as it's faster, to test correctness of the implementation instead of P-256, P-224 or P-192

[coveralls]

Coverage increased (+0.1%) to 98.699% when pulling 2273354 on tomato42:cosmic-ray into bbe3679 on warner:master.

[tomato42]

so, with those changes we end up with 222 test suites executed (6.7s per test suite execution), better, but not really good, that still means a 99.9% confidence interval of 10 percentage points

looks like we need to get down to around 1s per test suite execution to get workable confidence intervals

[tomato42]

CI run: https://travis-ci.org/github/warner/python-ecdsa/builds/747963586

[tomato42]

CI run: https://travis-ci.org/github/warner/python-ecdsa/builds/747987147

[tomato42]

CI run: https://travis-ci.org/github/warner/python-ecdsa/builds/748034897

[The-Mule]

Only nitpicks. I did no spotted any obvious flaw in the changes.

[The-Mule]

Nitpick: Isn't it easier (if possible) to use timeout 10m?

[tomato42]

good idea, but to implement it I would have to figure out error handling and so on... this is good enough (and probably easier to port to macos)

[The-Mule]

Nitpick: exception message sounds a bit odd

[tomato42]

that's because this isn't the whole message, the whole message is "Edwards keys don't support setting the secret scalar (exponent) directly"

[The-Mule]

Is it safe the second ImportError to be completely silent?

[tomato42]

why it wouldn't be? the point of this code is to define the mpz() method, where it comes from is secondary