Bump OpenCTM to fix heap overflows on crafted .ctm files (S9S disclosure)

[Fedr]

Summary

Bumps our OpenCTM dependency to the two fixes from the S9S Security Research coordinated disclosure, both reachable from ctmLoad() on a single crafted .ctm file and therefore from our own fromCtm load path (source/MRIOExtras/MRCtm.cpp), since the defects are in the library's allocation code, not in the read callback:

• Fix heap-buffer-overflow in _ctmStreamReadSTRING and stop parsing on string read errors OpenCTM#2 — heap-buffer-overflow write in _ctmStreamReadSTRING: malloc(len + 1) wraps for len = 0xFFFFFFFF, then up to len attacker bytes are read into a ~0-byte buffer (CWE-190 → CWE-787).
• Reject vertex/triangle counts whose allocation size would wrap 32 bits OpenCTM#3 — file-declared vertex/triangle counts drove malloc directly, so count * stride wrapped to an undersized buffer on 32-bit size_t (our wasm32 build) and in the 32-bit CTMuint products of the float-map and stream.c temp buffers. Counts above UINT_MAX / 16 are now rejected with CTM_BAD_FORMAT. Also frees a leaked MG1 temp buffer on a failed read.

Pinned to OpenCTM master at eae87a4d9ac229bdc7c32492c84bd7cb1329e5c4, the squash-merge of MeshInspector/OpenCTM#3 on top of MeshInspector/OpenCTM#2.

Changes

OpenCTM reaches the build three different ways, so all three pins move together:

• thirdparty/OpenCTM-git submodule (Linux dockers, macOS, emscripten) → eae87a4d9.
• thirdparty/vcpkg/ports/openctm overlay port (Windows, rockylinux8-vcpkg incl. manylinux wheels): REF + SHA512 → eae87a4d9's archive, port-version 3 → 4 (forces a rebuild of the cached binary).
• thirdparty/cpm/package-lock.cmake (CPM build from Add option to build third-party dependencies without submodules #6775) GIT_TAG → eae87a4d9.
• thirdparty/licenses/manifest.json re-pinned via check_third_party_licenses.py --update-versions.

Verification (in the OpenCTM PRs)

MeshInspector/OpenCTM#2: ASan build via the public ctmLoad() on the 44-byte reproducer: overflow before, clean rejection after; a valid mesh with a file comment still round-trips.
MeshInspector/OpenCTM#3: i686 (32-bit size_t) and x86-64 builds: a RAW file declaring 0x15555556 vertices segfaulted before and is rejected with CTM_BAD_FORMAT after; a valid RAW mesh still loads on both.