StrictRounding statement for floats

[cbmarini]

This PR introduces a new StrictRounding statement that rounds floats to a specified precision. If no precision is given, the default precision is used. This can be useful to trim the extra precision used by GMP. This also resolves #698.

Similar to the floating point precision, the precision in StrictRounding can be specified in digits or bits. For example:

• the base 10 number 1.1e-4 cannot be written as a finite sequence in base 2: it is 1.1100110101011111101*2^(-14). So, when we round to 5 bits, we get 1.1101*2^(-14) in bits. However, in base 10, this is written as 1.10626220703125e-04.

At the moment, the implementation makes use of the GMP functions mpf_get_str and mpf_set_str. Maybe in the feature we can come up with our own algorithm to do the rounding?

Example usage:

#-
Off statistics;
#StartFloat 6d
CFunction f;
Local F = f(1.0)+f(1.0000001);
L G = 1.23456789e-4;
Print;
.sort

StrictRounding 4d;
Argument f;
	StrictRounding;
EndArgument;
Print;
.end

results in

   F =
      f(1.0e+00) + f(1.0e+00);

   G =
      1.23457e-04;


   F =
      2*f(1.0e+00);

   G =
      1.235e-04;

[coveralls]

coverage: 56.79% (+0.01%) from 56.779%
when pulling 9088a07 on cbmarini:strict-rounding
into 2aba64b on form-dev:master.