Fix memory leak for merge_text_nodes

Recently introduced feature in lexbor backend did not free the data
untill the parser instance is alive.

Author: rushter

.gitignore

@@ -68,4 +68,6 @@ selectolax/lexbor.c
 
 # virtual env
 .venv/
-venv/
+venv/
+
+tmp/

CHANGES.md

@@ -4,6 +4,7 @@
 - Fix incorrect default value in docstrings for strict argument
 - Fix incorrect exception handling for `any_css_matches`
 - Fix docstring for `css_first` method
+- Fix memory leak in `merge_text_nodes` for lexbor backend
 
 
 ## Version 0.3.34

selectolax/lexbor/node.pxi

@@ -518,6 +518,10 @@ cdef class LexborNode:
                     combined = (<bytes>left_text[:left_length]) + (<bytes>right_text[:right_length])
                     lxb_dom_node_text_content_set(node, combined, len(combined))
                     lxb_dom_node_remove(node.prev)
+                    
+                lxb_dom_document_destroy_text_noi(self.node.owner_document, left_text)
+                lxb_dom_document_destroy_text_noi(self.node.owner_document, right_text)
+
             if node.first_child:
                 LexborNode.new(node, self.parser).merge_text_nodes()
             node = next_node