fix(lexer): don't split table cells on | inside links/code

The table-row lexer collected every `|` as a cell separator, so a pipe
inside a `[[url|title]]` wikilink (or `{{…}}` transclusion / `` `…` ``
inline code) was mistaken for a column boundary — mangling the link and
adding a phantom column in the AST and HTML export.

Replace the naive scan in `try_lex_table_row` with `cell_bar_positions`,
which skips pipes inside those constructs (matching the regions
`lex_inline` tokenises as a unit). Unclosed constructs fall back to
literal openers so a malformed cell can't swallow the rest of the row,
and the scan only slices at ASCII openers to stay UTF-8 safe.

Backslash-escaped `\|` is intentionally left unhandled: the inline lexer
has no escape handling, so honouring it here would leave a stray
backslash in the output. Documented as a matched follow-up.

Adds lexer + HTML-export regression tests.

Closes #1

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lzhq43v1qpkZczN65dqXa3

crates/nuwiki-core/tests/html_renderer.rs

@@ -245,6 +245,19 @@ fn table_with_header_separator() {
     assert!(out.contains("<td> 1 </td>"));
 }
 
+#[test]
+fn table_cell_with_wikilink_pipe_exports_one_cell() {
+    // Regression: the `|` in `[[url|title]]` must stay inside the link, so the
+    // row keeps two columns and the link renders as a single anchor. Before the
+    // fix the cell split into `[[https://example.com` + `My Site]]`, adding a
+    // phantom <td> and breaking the link.
+    let out = render("| Name | Link |\n|---|---|\n| foo | [[https://example.com|My Site]] |\n");
+    assert!(
+        out.contains("<td> foo </td><td> <a href=\"https://example.com\">My Site</a> </td>"),
+        "got: {out}"
+    );
+}
+
 #[test]
 fn single_comment_becomes_html_comment() {
     let out = render("%% hidden\n");

crates/nuwiki-core/tests/lexer.rs

@@ -454,6 +454,48 @@ fn table_col_and_row_span_cells() {
     );
 }
 
+#[test]
+fn table_cell_with_wikilink_pipe_is_not_split() {
+    // The `|` inside `[[url|title]]` is part of the link, not a cell
+    // boundary: the row must lex to exactly two cells (three TableSeps).
+    assert_eq!(
+        lex("| foo | [[https://example.com|My Site]] |\n"),
+        vec![
+            TableSep,
+            text(" foo "),
+            TableSep,
+            text(" "),
+            WikiLinkOpen,
+            text("https://example.com"),
+            WikiLinkSep,
+            text("My Site"),
+            WikiLinkClose,
+            text(" "),
+            TableSep,
+            Newline,
+        ]
+    );
+}
+
+#[test]
+fn table_cell_with_transclusion_and_code_pipes_are_not_split() {
+    // `|` inside `{{…}}` and inside `` `…` `` are both literal.
+    let lexed = lex("| {{a|b}} | `x|y` |\n");
+    let seps = lexed.iter().filter(|t| matches!(t, TableSep)).count();
+    assert_eq!(seps, 3, "expected 2 cells: {lexed:?}");
+}
+
+#[test]
+fn table_cell_with_unclosed_wikilink_falls_back_to_pipe_split() {
+    // A malformed, unclosed `[[` must not swallow the rest of the row —
+    // the pipes still delimit cells.
+    let seps = lex("| [[oops | tail |\n")
+        .iter()
+        .filter(|t| matches!(t, TableSep))
+        .count();
+    assert_eq!(seps, 3);
+}
+
 // ===== Definition list =====
 
 #[test]