fix(parser): preserve extra metadata safely

[denis-samatov]

Description

Motivation & Context

Changes Made

• [ ]

Verification & Pre-Submission Checklist

• make test passes with 100% test success
• make validate passes on knowledge/ (0 errors, 0 warnings, 0 orphans)
• make validate-examples passes on all sample corpora
• make fmt and make vet show clean code
• Any architectural or workflow changes are documented in knowledge/

[sknr]

Hi @denis-samatov,

Thanks for contributing these improvements! The changes to Unicode search, relation idempotency, and metadata round-tripping are well-designed and address real edge cases.

I have reviewed and merged your work directly into our integration branch develop (commits b7577f0, 49c1007, and e7263d9), and credited you as the author.

A quick note for future contributions:
Starting today, we have updated our contribution guidelines to an Issue-First workflow on the develop branch (see CONTRIBUTING.md):

Please open an Issue first to discuss ideas or bug reports before creating PRs.
Always base and target your PRs against develop (our default branch) rather than main (which is reserved for tagged production releases).
Please make sure to fill out the PR description template with context and verification details.
Thanks again for helping make okf-agent-memory better!