fix(renderer): fix table column width miscalculation in markdown.output()

[jujuyuzu]

Problem

Table borders (│) disappear on specific rows where cell content contains:

• Mid-word underscores: json_serializable, mock_template_repository
• Inline code with Lua pattern special chars (related cleanup)

Reproduction

| Item                                                                     | Difficulty |
| ------------------------------------------------------------------------ | ---------- |
| Extract AccordionCard from workout detail + template detail              | Medium     |
| `catch (_)` → `firstWhereOrNull` in mock_template_repository                 | Easy       |
| Remove unused pubspec deps (json_serializable, riverpod_generator, etc.) | Easy       |
| SetHeaderRow: add `readOnly` param to hide dropdown                        | Easy       |

Rows 2 and 3 lose their left/middle borders. Rows 1 and 4 render correctly.

Root Cause

Three bugs in markdown.output() (used for table column width calculation):

1. Overly aggressive italic _ matching (line 244)

%_(.-)%_ matches mid-word underscores in identifiers like
json_serializable. CommonMark/GFM spec says _ emphasis requires word
boundaries. Treesitter correctly ignores mid-word _, but
markdown.output() strips them → width mismatch → broken borders.

2. corner_left typo (8 occurrences)

corner_left used where corner_right was intended — the last element
in concat() calls after padding_right. Affects inline code, highlights,
email, and URI handlers.

3. Redundant vim.pesc() inside concat()

concat() (line 27-33) already calls vim.pesc() on every element.
Three call sites additionally wrapped arguments with vim.pesc() before
passing to concat(), causing double-escaping and potential match failures.

Fix

• Add word-boundary check for italic _: skip when _ is preceded or
  followed by %w (alphanumeric), matching CommonMark rules
• Fix corner_left → corner_right in all 8 affected locations
• Remove redundant vim.pesc() calls in 3 locations

[OXY2DEV]

@jujuyuzu thanks for the commit.

The original pattern approach was error-prone, slow and had portability issues. I have re-written it to address all of these issues(including the bug mentioned in this commit).

Give it a try.