The subtype of a binary value is a std::uint64_t, but dump() wrote it
with dump_integer, which converts its argument to the magnitude type of
number_unsigned_t. With a number_unsigned_t narrower than 64 bits, the
subtype was truncated (subtype 300 with std::uint8_t numbers was
printed as 44).
A dump_subtype helper now writes the subtype for both the compact and
the pretty-printed output: through dump_integer if the magnitude type
of number_unsigned_t has at least 64 bits (the default json, so its
output and speed are unchanged), and with a digit loop otherwise.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>