Fixes #3096, and part of of GNU's tests/od/od-float.

Some TODOs, for follow-up PR (I'll do some of it and file issues for the rest):

• Performance is bad -- some of that is expected, but for example, we don't need all 100 digits of the default BigDecimal precision when doing the exponentiation.
• I feel that on arm64, we should actually parse a f128 (IEEE 754-2008 binary128), and maybe fall back to f64 on arm.
• prn_float.rs should also use uucore's formatting functions (that code is quite old), that'd make the printing a bit "better", i.e. just print 1 vs 1.000000. => that'll fix tests/od/od-float I think.
• We don't exactly handle printing like GNU, but I think this might not matter too much (GNU seems to trim digits if it still lands unambiguously on the same binary representation).

test_od: Add f128 test (actually a f80 test)

od: Add printer for ExtendedBigDecimal (for f80/f128)

od: input_decoder: f80 parsing code for -t fL format

This replicates x86(-64) behaviour.

od: output_info: Add support for 16 byte long output

fL can consume 16 bytes at a time, so we need to be able to
generate correct padding.

Also add a simple test for such wider lines.

od: Start adding support for f128 (ExtendedBigDecimal-backed)

There is no (stable) support for f128 (yet?), and we need to parse
f80 anyway, so let's back the datatype with an ExtendedBigDecimal,
for now.

uucore: extendedbigdecimal: Add from_number_exp2 to compute num * 2**exp

Move the code from num_parser, as we have another usecase in od.