_escape_ch writes NUL and vertical tab as \0 and \v. Those escapes exist in JSON5 (via ECMAScript) but not in JSON, where the only single-character escapes are \", \\, \/, \b, \f, \n, \r and \t. dumps(quote_keys=True, trailing_commas=False) is documented to produce the same output as json.dumps(), and --as-json in the CLI uses those settings, but for any string containing U+0000 or U+000B the result is "\0" or "\v", which json.loads() and other strict JSON parsers reject with Invalid \escape. I noticed it while comparing dumps() against the json module for every BMP code point; these two were the only characters whose output could not be read back as JSON. Letting them fall through to the generic \uXXXX form gives the same six-character \uXXXX escapes for NUL and VT (\u000b for VT), which is what the json module writes and is equally valid JSON5, so loads() round-trips are unchanged. That also makes the NUL-before-digit special case in _encode_quoted_str from #121 unnecessary, since a \uXXXX escape for NUL is never ambiguous, so it is removed.

>>> import json, json5
>>> json5.dumps('a\0b\v', quote_keys=True, trailing_commas=False)
'"a\\0b\\v"'
>>> json.loads(_)
json.decoder.JSONDecodeError: Invalid \escape: line 1 column 3 (char 2)