Add Playfair cipher codec

The n == 10 fast-path returned through str(int)/int(str)/i2s before reaching the
generic divmod path, so base10 (and its int/integer/dec/decimal aliases) both
missed the leading-zero handling added for the other big-integer bases and had
two more latent round-trip bugs:
- leading null bytes were dropped, e.g. b'\x00abc' -> '6382179' -> b'abc';
- i2s() does hex(i)[2:].rstrip("eL"), so any value whose integer ends in a 0x?e
  byte lost that nibble on decode: b'd\xc6\xfe' -> '6604542' -> b'\x06Lo';
- str()/int() hit CPython's integer string-conversion digit limit, raising
  ValueError on inputs larger than ~1780 bytes.

Keep the fast-path only for a non-standard 10-character charset and let the
digits charset use the generic loop, which already handles all three cases.

The width-16 and crc-a variant tables were the only ones without a ''
default entry, so encode(data, "crc16") and bare "crc" raised
KeyError: '' instead of resolving like every other width. Default them
to CRC-16/ARC and CRC-A respectively (both already present as named
aliases).

Fixed KeyError on bare crc16 and crc codecs

Fix base10 dropping leading zeros and corrupting some values on decode