fix(xclip): follow-ups to the XCLIP merge

- An inverted clip's roundtrip record goes in the filter's extension
  dictionary through the codec's own API (ensure_extension_dictionary /
  xrecord / extension_dictionary_handle) instead of a scan of every
  object, which the renderer ran for each clipped reference.
- Pin cadcodec 1c3be0c (the DXF writer now links a spatial filter's
  extension dictionary, so that record is not left unreferenced) and
  cadkernel deb529b; plugin crates and the Python plugin lock follow.
- The XCLIP boundary preview stays f64.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/cadcodec-reader-gaps.md

@@ -7,7 +7,7 @@ reloading it, and comparing entity fields.
 **Status (23 September 2026):** the fixes landed in cadcodec as
 [#48](https://github.com/HakanSeven12/cadcodec/pull/48) (issues 1-6) and
 [#51](https://github.com/HakanSeven12/cadcodec/pull/51) (the first three style findings), and
-the block description with commit `dd1d7bf` (cadcodec issue #49). OCS pins `d290dbe`, which
+the block description with commit `dd1d7bf` (cadcodec issue #49). OCS pins `1c3be0c`, which
 includes it; the canaries below were flipped against `dd1d7bf`. What is still open:
 
 - **ATTDEF `lock_position`** (issue 1): the reader is complete, but the DXF writer emits no