Support randmatrix() for mp.iv and mp contexts

[MaxGaukler]

Before, mpmath.iv.randmatrix() caused an error. Now it works correctly:

>>> import mpmath
>>> mpmath.iv.randmatrix(3)
matrix(
[['[0.66382280562735496154, 0.66382280562735496154]', '[0.016861908353603016764, 0.016861908353603016764]', '[0.73734056129022396142, 0.73734056129022396142]'],
 ['[0.36647301100616180847, 0.36647301100616180847]', '[0.62220372249504873885, 0.62220372249504873885]', '[0.48761902415903068952, 0.48761902415903068952]'],
 ['[0.81322606973113087392, 0.81322606973113087392]', '[0.15903271707414323721, 0.15903271707414323721]', '[0.090983058161586249923, 0.090983058161586249923]']])
>>> type(mpmath.fp.randmatrix(3)[0,0])
<type 'float'>
>>> type(mpmath.mp.randmatrix(3)[0,0])
<class 'mpmath.ctx_mp_python.mpf'>
>>> type(mpmath.iv.randmatrix(3)[0,0])
<class 'mpmath.ctx_iv.ivmpf'>

• Is it okay that I moved import random to the top level? AFAIK it should even be faster this way.

[fredrik-johansson]

There is a small problem: random.random() returns a random double, so this will not give full precision random numbers.

[mgmax]

Should be better now. I added tests, but am not sure how workprec() is supposed to work for iv. Is that a bug or by design?

[MaxGaukler]

By the way, please let me know as soon as you consider it ready for merging so that I can provide a rebased version of this pull request as a single commit.

[skirpichev]

@MaxGaukler could you please fix merge conflicts?

[MaxGaukler]

Now squashed and rebased. Happy to see that there is interest in my suggestion :-)

[skirpichev]

There are also merge conflicts. Please fix them.

Except for comment in the ctx_fp.py and above remark - seems ok for me.