Add half-precision type with higher maximum value.

Adds the option to write half-precision HDF5 output with the exponent bias lowered by 10.

Unfortunately the exponent in IEEE half-precision floats overflows around 65k. This isn't always ideal since we store energies and masses in MeV: you can't even store the mass of the Higgs in our units at half precision!

Fortunately we can just lower the exponent bias by 10, and store numbers 1024 times larger. That gets our upper bound for mass and energy up around 65 TeV (which we probably won't see a whole lot of at ATLAS) at the expense of sub-MeV energies.