[TS] Preserve sparse array slots in source mutation models

[CaelmBleidd]

The reverse and unshift source models rejected arrays with known holes, stopping previously supported sparse-array programs. Extend their accepted receivers to arrays with known own-element presence or a dense-array proof. The TypeScript bodies now preserve missing elements through ordinary indexed writes and deletes; model-owned prepareOwnSlots and hasOwnSlot primitives provide the required metadata.

All calls continue through normal model selection, application events, and residual fallback. The generic call resolver and dispatcher have no special sparse-array handling. The models retain their length/result limit of 16 and the existing unshift arity bound of three arguments.

Regressions cover asymmetric holes, aliases, boxed values, conditional deletion, zero/multiple unshift arguments, disabled models, and Node.js replay. They explicitly require ModelApplied events. Three of the six sparse regression tests fail on the previous PR head because its resolver bypass does not apply the source models.

Validation on the final implementation:

• Full :usvm-ts:check :usvm-ts-dataflow:check :usvm-ts-calls:check passed locally: 1510 passed, 175 skipped, zero failures/errors.
• JDK 11, ETS_IR_PROVIDER=ts-frontend, sequential execution; MockK JVM self-attach enabled locally.
• Detekt main/test and git diff --check passed.
• Native Node.js replay passed for the sparse source witnesses.

Remote CI must validate the published head before merging.