[sandsand244263]

Adds manifest crypto support for build 2.25.0.1.154504N.

• ProCMF_154504 — CMF procedure for 154504. Same keytable as before is not reused: this build derives key/iv differently (kidx = T[0x100 + len] with += 3 stepping for the key, and a dataCount/digest-driven step for the iv), so the table for this build is embedded as a const array like the existing ProCMF_* files.
• ProTRG_154504 — TRG procedure for 154504, registered via AutoDetectVersion. Shares the same keytable and key derivation as the CMF procedure; only the iv derivation differs (uses the header's package count instead of the manifest data count).

Verified against the real client manifests (build 154504):

• CMF Win_SPWin_RCN_EExt.cmf: 361 entries, 1,306,039 hashes — entry indices unique/increasing starting at 1, hashes ascending for all 1,306,039, all sizes sane, plaintext length exactly 361*20 + 1,306,039*29 + 1.
• Language manifests (..._LzhCN_text_EExt.cmf, ..._LzhCN_speech_EExt.cmf) decrypt with the same table; 124,183 and 66,220 hashes respectively, all GUIDs strictly increasing, padding bytes consistent.
• TRG: 49 MB blob with matching tail padding; packageCount * 30 == packageBlockSize; walking all 15,584 skins via m_assetPtr lands exactly on the block end (no trailing free bytes); 95.8% of the 29,516 packages have resourceKeyID == 0.
• End to end: list-workshop on this build produces byte-identical output to the previous build's export, i.e. this is a pure build bump with no data changes.

Notes:

• One uncertainty worth flagging: the TRG iv's packageCount term is behaviourally inferred (it is the only candidate that makes step land correctly), not read off a disassembly operand. It only affects the first 16 bytes of plaintext, and the structural checks above pass with it.
• AttemptFallbackManifests is left alone; these procedures register normally.
• The WinPrism_* product branch is untouched (DataTool filters it out on this platform anyway).

[neptuwunium]

Hi
Thanks for your AI generated pull request.
Regards,
ada