Fix O(n^2) parsing time with --fspoilers and deeply nested brackets

The spoiler analysis loop in md_analyze_link_contents() was called
recursively for every resolved link pair (by md_resolve_links()), but
lacked the skip-resolved-spans optimization present in md_analyze_marks().

With deeply nested bracket structures (e.g. the OSS-Fuzz testcase
clusterfuzz-testcase-fuzz-mdhtml-5579634954797056 with 37,034 nesting
levels), the loop iterated O(n^2) over the same marks, causing 13+
second parse times for a 661 kB input.

Fix: when the loop encounters a resolved link or image opener ('[', '!'),
jump directly to mark->next (the closer) instead of walking through
every mark inside the span one by one. This mirrors the skip logic in
md_analyze_marks() and reduces total iterations from O(n^2) to O(n).

Closes #311.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: hryhoriiK97

src/md4c.c

@@ -4173,8 +4173,15 @@ md_analyze_link_contents(MD_CTX* ctx, const MD_LINE* lines, MD_SIZE n_lines,
     if(ctx->parser.flags & MD_FLAG_SPOILERS) {
         for(i = mark_beg; i < mark_end; i++) {
             MD_MARK* mark = &ctx->marks[i];
-            if(mark->flags & MD_MARK_RESOLVED)
+            if(mark->flags & MD_MARK_RESOLVED) {
+                /* Skip resolved link/image spans. Their contents were already
+                 * analyzed by the recursive md_analyze_link_contents() call
+                 * inside md_resolve_links(). Without this, deeply nested
+                 * brackets cause O(n^2) parsing time. */
+                if((mark->flags & MD_MARK_OPENER)  &&  ISANYOF_(mark->ch, "[!"))
+                    i = mark->next;
                 continue;
+            }
             if(mark->ch == '|' && mark->end - mark->beg == 2)
                 md_analyze_pipe(ctx, i);
         }