Fix use-after-free of the clothes TXD buffer after engineImportTXD

[Flashmyname]

Summary

AddClothingTexture hands m_FileData.data() to ClothesAddFile, which stores the raw pointer (ClothesReplacing.cpp:164), but never marks the buffer as borrowed. A later engineImportTXD on a normal model frees it at CClientTXD.cpp:135.

The flag is now set when ClothesAddFile succeeds, and the three bools in CClientTXD get default initialisers.

Motivation

The streaming hook finds the freed pointer by name (ClothesReplacing.cpp:267) and passes it to the model loader (:286). ClothesRemoveFile matches by pointer (:186), so the destructor cleans up nothing and the name stays registered - every later engineAddClothingTXD with it fails with "Texture already added".

m_bUsingFileDataForClothes was also read at :134 before anything wrote it - the constructor sets only m_pManager - so the failure was build dependent.

Two intended consequences: the raw bytes now live until the element is destroyed, which is the contract :104-106 already imposes, and a clothes import after a non-clothes import now works instead of failing at :84.

CClientDFF looks the same on its raw data path: AddClothingModel registers m_RawDataBuffer.data() (CClientDFF.cpp:116), GetLoadedClump frees the buffer at :66, and RestoreModels then deregisters a pointer that no longer matches (:253). I haven't tested that path, and fixing it means touching the free that limits a raw data DFF to a single replace, so I've left it out of this one.

Test plan

Three cycles of engineLoadTXD, engineAddClothingTXD with a fresh name, optionally engineImportTXD(txd, 1337), then destroyElement - one cycle per frame, since destroyElement only queues the element for CElementDeleter.

                          cycle 1  2       3
before  without import    ok       ok      ok
        with import       ok       FAILED  FAILED
after   without import    ok       ok      ok
        with import       ok       ok      ok

Checklist

• Your code should follow the coding guidelines.
• Smaller pull requests are easier to review. If your pull request is beefy, your pull request should be reviewable commit-by-commit.