Float() and numer



Currently, float(2^%pi) returns 2.0^%pi, and there is no way to persuade it
to return 8.82 -- except something ridiculous like scanmap(float,2^%pi).  On
the other hand, ev(2^%pi,numer) returns 8.82.

This is because FLOAT has an explicit special case for ^ and ^^, where it
does not evaluate the exponent as a float.  This is to avoid things like
float(x^2-x+1) => x^2.0-x+1.

I propose that the NUMER flag apply not only to MEVAL, but also to FLOAT, so
that block([numer:true],float(2^%pi)) => 8.82... but also
block([numer:true],float(x^2)) => x^2.0.  I believe this makes it easier to
calculate the numeric value of an expression without evaluating it using EV,
which is often inappropriate (cf. bug report 1724592).

This doesn't break anything in the testsuite (not surprising).  Any
objections?

               -s