Use init() to set a function pointer to the fastest implementation
of fp751MontgomeryReduce based on CPU capabilities.

In https://eprint.iacr.org/2017/1015.pdf a technique was described to
improve the performance of Montgomery reduction for Montgomery-friendly
moduli. This adds an implementation using the mulx, adox and adcx
instructions, available in the BMI2 (since Haswell) and ADX (since
Broadwell) instruction set extensions.

* Code is much slower than x86 specialized implementation
* Cross checked on ARMv8 (32-bit)

* implements SIKE specified here:
    http://www.sike.org/files/SIDH-spec.pdf
  * methods for both - KEM and PKE - are added
  * adds SIKE specific key variant
  * tests: known answer tests for sike
  * uses cSHKAE from nobs-crypto

* tests: adds continues integration

* Makefile has targets for running tests, benchmarks and code coverage. It also
  contains target for env preparation. In order to run sidh tests
  GOPATH must contain p751toolbox package. I've chosen to manualy
  copy this package to the temporary GOPATH directory. It could also be done
  by calling "go get", but then any commit to both p751toolbox and sidh would need
  to be done in 2 steps.
* .travis.yml calls make and uploads code coverage to Codecov

* move sidh to seperated folder

* sidh: updates algorithm to SIDHv3 and refactoring

* makes an algorithm compatible with Microsoft's SIDHv3
  implementation. This is required to implement SIKE key
  encapsulation mechanism, as specified in PQC NIST submission
  from Nov, 30 2017
* removes SIDHBobPublicKey/SIDHAlicePublicKey/SIDHBobPrivateKey/
  SIDHAlicePrivateKey. Instead PrivateKey and PublicKey structures
  where introduced. Each of this structure stores variant of a key
  A or B. Implementation uses a key variant in order to differentiate
  between 2- and 3-torision groups.
  Main goal of removing "Alice" and "Bob" specific structures is to
  remove code duplication
* Introduces SidhParams: structure to store prime field and SIDH
  domain parameters.
* Refactors public API. Introduces:
    - Functions to generate, import, export keypair
    - DeriveSecret function to create shared secret
    - Supporting functions and types
* Removes code which is not used by implementation anymore, like
  DistortAndDifference(), SecretPoint(), DblAdd(),
  OkeyaSakuraiCoordinateRecovery() and many more. Also tests for those
  functions are removed.
* Adds fixes for key import/export

Optimizations derived from armfazh/flor-sidh-x64 project.

- optimizations.md file removed.
 - DblAdd function now is a method of ProjectivePoint and ProjectivePrimeFieldPoint.
 - A better (re)utilization of variables inside of DblAdd method.
 - Reflect package is not required anymore in p751toolbox/print.go.

Look at optimizations.md file.

Use "shared secret" instead of "key agreement" to point out that there's no hashing, key derivation, etc. going on.

Measure the keypair generation (including random sampling) in the "fast" benchmarks.

A later revision of the CLN paper uses the Montgomery relation to recover the a
value from the three x-coordinates in the public key, saving its transmission.
This patch improves slightly over the description in the paper, by
projectivizing the computation of the first four-isogeny; this means that it's
not necessary to perform an inversion to get the affine a value.