fix: preserve literal HTML in Markdown table exports

[lllleolin-max]

Description

Copying or downloading a table as Markdown can change its cell text when the export is rendered again. For example, Array<string> becomes Array, and the literal text © becomes ©.

Escape &, <, and > in the shared Markdown cell serializer so those values survive export. The existing single-pass serializer still emits <br> for actual newlines.

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Changes Made

• Preserve literal HTML and character entities in Markdown table exports, including headers.
• Add a DOM → Markdown → DOM regression test using the existing marked dependency.
• Add a patch changeset for streamdown.

Testing

• Added new tests for the changes
• All existing tests in the changed package pass: pnpm --filter streamdown test — 85 files, 1,173 tests.
• Confirmed the new regression fails on the original serializer and passes with the fix (47 table utility tests).
• pnpm -r --filter 'streamdown...' build — core and dependency packages build, including declarations.
• pnpm check — 291 files checked.
• Manually tested the changes

An extra pnpm --filter streamdown exec tsc --noEmit check reports 211 existing diagnostics on both the original and patched source; diagnostics match after normalizing source positions. This check is not passing on the baseline either.

Checklist

• My code follows the project's code style
• I have performed a self-review of my own code
• I have added tests that prove my fix is effective
• New and existing unit tests in the changed package pass locally
• I have created a changeset

Changeset

• I have created a changeset for these changes

Additional Notes

Prepared with OpenAI Codex. Automated tests and Codex agent review were completed; no independent human review is claimed.

[vercel]

@lllleolin-max is attempting to deploy a commit to the Vercel Team on Vercel.

A member of the Team first needs to authorize it.

[farnabaz]

Thanks