[codex] Harden Windows memory dump parsing

[tianmind-studio]

Summary

• Skip Markdown code fence lines when parsing pasted ChatGPT/Claude memory dumps.
• Strip additional common pasted list markers, including em dash, triangular bullet, and hyphen bullet variants.
• Add focused coverage for fenced dumps and richer pasted list markers.
• Format the touched fence test with the current Windows app dev dependencies.

Why

Windows users often copy memory lists from chat surfaces that wrap content in Markdown fences or use typographic bullets/dashes. The existing parser could keep ``` fences as bogus memories and leave some common marker characters in the imported text.

Refresh

• Rebased on main at b781d91767ac96a20f2305614eb50c7744a8a6c3.
• Current head: 22e3d0ca67e4c7afe4141ad3fb919f1d7dec5f89.
• GitHub currently reports this PR as mergeable.

Validation

• From desktop/windows/: npm run test -- src/main/memoryImport/parse.test.ts -> 1 file, 8 tests passed
• From desktop/windows/: npm run typecheck -> passed
• targeted npx eslint --quiet on changed Windows files -> no errors
• git diff --check origin/main...HEAD -> passed
• git diff --check -> passed
• scripts/pre-commit with PYTHONUTF8=1 -> passed
• scripts/pre-push with PYTHONUTF8=1 -> passed

[Git-on-my-level]

Thanks for tightening up the Windows memory import parser. This is a focused, low-risk improvement: it correctly skips Markdown fence lines, expands support for common pasted bullet markers, and includes targeted Vitest coverage for both behaviors. The existing lint/format check is passing, and I did not see any security, dependency, workflow, or product-scope concerns.

[kodjima33]

Desktop bug fix (owner-override): harden Windows memory-dump parsing (skip code fences, more list markers).