Eensy-weensy teeny toy_multisplit optimization.

For every position in the string, we shave off a substring
of the longest possible separator.  Then we clip off progressively
shorter substrings, until we find a substring that matches one
of the separators of that length.  If the string doesn't start with
any separator, we'll try 'em all.

I happen to know that CPython has some hacks that speed up string
manipulation.  If you only have one reference to a string, and
you transform the string somehow and overwrite your reference with
the transformed verison, CPython can often simply modify your
existing string in-place.

So, I rewrote toy_multisplit so it pre-creates the transformed
string, then shaves characters off *that*.  And it does appear
to be a teeny tiny bit faster.

tests/test_text.py

@@ -1124,9 +1124,11 @@ def flush_word():
                 segments.append(empty.join(word))
                 word.clear()
 
+            longest_separator_length = separators_by_length[0][0]
             while s:
+                substring = s
                 for length, separators_set in separators_by_length:
-                    substring = s[:length]
+                    substring = substring[:length]
                     # print(f"substring={substring!r} separators_set={separators_set!r}")
                     if substring in separators_set:
                         flush_word()