Implement tensor access logic

[code-tc]

Summary

Implement and validate the full non-transactional access layer for `fensor` across dense and sparse layouts, including view semantics, sparse iteration order constraints, lifecycle policy for sparse zero writes, and fail-closed persistence behavior.

Scope

• Implement/complete:
  • `TensorRead`
  • `TensorWrite`
  • `TensorReadBulk`
  • `TensorWriteBulk`
  • `TensorTransform` (`reshape`, `slice`, `transpose`, `broadcast`, `flip`, `squeeze`, `unsqueeze`)
  • `TensorBlockStore`
  • `TensorSparseIndex`
  • `TensorViewSemantics`
  • `TensorSparseLifecycle`
• Finalize `read_sparse_elements_in_order` semantics:
  • Support valid in-order sparse iteration paths
  • Return structured `UnsupportedSparseIterationOrder` on incompatible order requests
• Persist/rehydrate semantics:
  • Confirm create/write/reload/read flows
  • Corruption handling must fail closed (no auto-recovery behavior)

Explicit Non-Goals

• LSM transactional orchestration (`pending/committed/finalize`) in `tc-collection`
• Full math-op implementation (`TensorMath*`, `TensorReduce*`, `TensorMatMul`) beyond access/transform prerequisites

Acceptance Criteria

• Dense/sparse access parity tests pass for supported operations
• Base/view contract is explicit and tested:
  • Which tensors are writable
  • Which views support write-through vs read-only behavior
• Sparse zero-write lifecycle policy is implemented and tested:
  • nonzero -> zero
  • zero -> nonzero
  • index/block state transitions are deterministic
• `read_sparse_elements_in_order`:
  • Works for supported orderings
  • Returns structured error for incompatible ordering with actionable message
• Persistence integration tests pass using real `freqfs` reload boundaries
• Corrupted metadata/data fails with structured errors (fail-closed)
• No `todo!`, `unimplemented!`, or panic-based placeholders in production access paths

Test Plan

• Add/extend tests in:
  • `deps/fensor/tests/foundations.rs`
  • integration tests for persistence/reload/corruption
• Include:
  • dense vs sparse parity matrix for access+transform
  • chained transforms (`transpose(slice(x))`, `slice(transpose(x))`, etc.)
  • sparse order compatibility/incompatibility cases