fix(agent_hub): deliver memory in each target framework's native runtime format

[mushenL]

Change Summary

Loose memory detail now lands where each target's runtime actually reads
it: hermes gets §-delimited entry stores (ported from its official
importer: extract, dedupe, char-budget skip so its drift guard passes),
nanobot/openhuman inline into their single memory file (openhuman under
its 2000-char injection cap), openclaw uses its own memory/imports//
convention. Index links are rewritten alongside the moves so no dangling
indexes remain, and qoder targets get index references appended. Drop the
unsubstantiated qwenpaw/qoder USER.md mappings (dead files) in favor of
catch-all folding; render hermes § stores as Markdown when it is the
source; report non-Markdown memory payloads explicitly instead of a
generic drop line. openhuman: collect config.toml from its real location
one level above the workspace (activating the TOML scrubber that never
ran), treat blank profile files as absent per the app's own fallback
semantics, collect MEMORY_GOALS.md, and stop collecting the derived wiki
vault.

Checklist

• The pull request title is a good summary of the changes - it will be used in the changelog
• Unit tests for the changes exist
• Run pre-commit install and pre-commit run --all-files before git commit, and passed lint check.
• Documentation reflects the changes where applicable