Fix realloc in retokenizer.split() (explosion#4606)

Always realloc to a size larger than `doc.max_length` in
`retokenizer.split()` (or cymem will throw errors).

spacy/tests/doc/test_retokenize_split.py

@@ -183,3 +183,18 @@ def test_doc_retokenizer_split_lex_attrs(en_vocab):
         retokenizer.split(doc[0], ["Los", "Angeles"], heads, attrs=attrs)
     assert doc[0].is_stop
     assert not doc[1].is_stop
+
+
+def test_doc_retokenizer_realloc(en_vocab):
+    """#4604: realloc correctly when new tokens outnumber original tokens"""
+    text = "Hyperglycemic adverse events following antipsychotic drug administration in the"
+    doc = Doc(en_vocab, words=text.split()[:-1])
+    with doc.retokenize() as retokenizer:
+        token = doc[0]
+        heads = [(token, 0)] * len(token)
+        retokenizer.split(doc[token.i], list(token.text), heads=heads)
+    doc = Doc(en_vocab, words=text.split())
+    with doc.retokenize() as retokenizer:
+        token = doc[0]
+        heads = [(token, 0)] * len(token)
+        retokenizer.split(doc[token.i], list(token.text), heads=heads)

spacy/tokens/_retokenize.pyx

@@ -329,7 +329,7 @@ def _split(Doc doc, int token_index, orths, heads, attrs):
             doc.c[i].head += offset
     # Double doc.c max_length if necessary (until big enough for all new tokens)
     while doc.length + nb_subtokens - 1 >= doc.max_length:
-        doc._realloc(doc.length * 2)
+        doc._realloc(doc.max_length * 2)
     # Move tokens after the split to create space for the new tokens
     doc.length = len(doc) + nb_subtokens -1
     to_process_tensor = (doc.tensor is not None and doc.tensor.size != 0)