Fix encode_buf when a maximal-length run is immediately followed by a zero

[jgallagher]

Prior to this change, encode_buf() would erroneously omit an encoded zero byte if that zero byte was immediately preceded by a MAX_LEN-length run. In addition to a fix, this PR adds:

• A test fixture with a buffer that triggers this case
• A fuzz test comparing encode_buf against encode_iter (which handled this case correctly)
• A fuzz test that round trips through encode_buf and decode_in_place

[cbiffle]

Nice catch.