Add batch support, improve random generation, improve multiprecision trig.

[paulcaprioli]

This PR consists of three commits:

• Add support to benchmark a batch of FFTs.
• Replace system drand48() with portable C code RNG.
• Improved accuracy of multiprecision trig.

Note that batch support was added to libbench and to the fftw3 and intel-mkl benchees.
Note that floating point random numbers are generated by directly setting significand bits from a uniform distribution. The bit generator is xoshiro256plus. (Note that doubles have more than 48 bits to set.)
Note that the accuracy of the multiprecision trig functions is improved by roughly 10 bits and also is slightly faster.
Also, in that file, REAL is defined to be double, to match the output precision of fftaccuracy() as well as the definition of BITS_IN_REAL.
Please feel free to modify the copyright notice at the top of the file as you wish. The GPL says I should add my name, but there was nothing there....