Finish Number position, resource and particle interfaces after #2744

[andrewboudreau]

Retain the remaining Number source reconstruction work after the PR #2744 correctness repair.

Next owner: codex-integrator-0918 (root coordinator). Finding: NUMBER-2744-05. This issue records a bounded continuation; it does not establish Source review acceptance.

The current local repair uses ordinary velocity, height and state updates, real Vector3 storage, and no volatile or integer-pointer scheduling barriers. Producer probes verify all eight Number functions and relocation destinations under 2004/b56; independent acceptance is pending. Three incorrect local call contracts are repaired and are not deferred here.

Remaining work:

• Replace the two Vector3 views over the base actor's scalar position members through a separately owned reconstruction of the shared position layout, if supported by consumer evidence. In the measured Number forms, direct particle Y or direct owner scalar reads produce a 368-byte Behavior instead of 372; a named whole-position reference produces 376 bytes. The present casts are a retained source compromise, not proof that ordinary C++ cannot match.
• Recover SharedFilePtr's loaded-file member and replace its +4 views. Reusing LoadFile return values in this TU produced a 436-byte InitResources instead of 444. The shared class currently exposes no recovered data fields.
• Reconcile the shared TextureSequence SetFile declaration with its implemented u16 final argument, then investigate a genuine member call. The corrected Number bridge agrees with the actual definition. A probe using the existing shared member and a local Fix12 speed produced 468 bytes instead of 444; that existing header still declares the final argument u32, so the probe is not a proposed contract fix.
• Reconstruct Particle::System::NewSimple in its actual shared class and verify its consumers before replacing Number's scalar bridge. The bridge now correctly returns void*; the current shared header has no NewSimple member. No compiler impossibility is claimed.
• Recover the PowerStar halfword at 0x496 before replacing the remaining raw read. Number now uses the existing PowerStar::unk_43c member for the other condition. Offset 0x496 is still inside unrecovered padding.

Reproduction evidence will accompany the repair in notes/experiments/number-2744-source-forms.md. Start from the final reviewed PR #2744 repair, preserve attribution and actual callee contracts, and require independent source review plus strict object/relocation, affected-consumer and full-ROM proof for any change.