Fix _dump_str crash and wrong encoding for hex escape sequences

[bysiber]

Summary

_dump_str in encoder.py crashes with an IndexError when encoding strings that start with a control character (like \x01), and silently produces invalid TOML for strings with control characters in the middle.

Problem

The \x escape handling in _dump_str uses a split-and-rejoin approach that has two issues:

1. When the \x escape occurs at the very start of the string, split("\\x") produces an empty first element. The code removes this empty element with v = v[1:], but then unconditionally accesses v[1] which no longer exists → IndexError.

2. The joiner logic is inverted. When the character before \x is NOT a backslash (meaning it's a real hex escape from repr()), the code uses "x" as the joiner — which just drops the backslash and puts the literal character x back. This produces wrong output like "ax01b" instead of "a\u0001b".

>>> import toml
>>> toml.dumps({"key": "\x01hello"})
# IndexError: list index out of range

>>> toml.dumps({"key": "a\x01b"})
'key = "ax01b"\n'
# Expected: 'key = "a\\u0001b"\n'

>>> toml.loads(toml.dumps({"key": "a\x01b"}))
{"key": "ax01b"}
# Roundtrip is broken — data is corrupted

Fix

Replace the fragile split/loop approach with a straightforward character-by-character scan that processes backslash sequences left-to-right:

• \\ (escaped backslash pair) → kept as-is
• \x (hex escape) → converted to \u00 for TOML compatibility
• Other \X sequences → kept as-is

This correctly handles all combinations including \\x (literal backslash + x), \\\x (literal backslash + hex escape), and strings starting with \x.