Accept floating-point NaN and Inf as fnumber

[rwpeterson]

I was working on parsing numerical data and found that none of pyparsing's numerical expressions support the non-finite floating point values (NaN, +Inf, and -Inf).

Just based on the implementation, ppc.fnumber looks like a reasonable place to add support for this, since it's independent of the other numerical expressions and has the float conversion baked in. (Indeed, I initially thought it would accept these values due to the "float" part.) Essentially, this PR makes fnumber match anything that Python's float can itself parse. cPython's parsing is fully case-insensitive and accepts (optionally signed) "NaN", "Inf", and "Infinity".

One note of caution would be if users somehow rely on fnumber's failure to parse such constants as a way to deal with NaNs or infinities in their programs.

[ptmcg]

To avoid introducing a regression on fnumber, what do you think about adding separate expressions for nan's and inf's, so that apps that need them could use an expression like numeric = fnumber | nan | inf? Or create another version of fnumber that adds these - just need to come up with a decent name (one of the 2 hardest problems in CS).

[rwpeterson]

I agree on keeping it separate from fnumber, and on the difficulty of a name :)

I factored it out into a separate ieee_float, since NaNs in particular are associated with the standard. Some other options for the name could be float_literal, ieee754, etc.

[ptmcg]

ieee_float looks good to me, and I admit, I never would have come up with that. Thanks for submitting!