Add io/read state machine for copy-based tensor loading

[gabewillen]

Summary

Add an io/read state machine under src/emel/io that provides an explicit read/copy tensor-loading strategy beneath tensor-owned residency management.

This issue should land after #59 and #60.

Motivation

Not every platform or deployment mode should rely on memory mapping. We need a separate strategy for explicit file reads / copies into owned buffers.

That strategy has distinct behavior:

• owned-buffer residency instead of mapped residency
• explicit transfer/copy lifecycle
• different error and resource behavior than mmap

It should be implemented as its own io actor rather than folded into tensor orchestration or bundled with other strategies.

Scope

• Add src/emel/io/... machine(s) for explicit read/copy tensor loading.
• Define the request/result events needed for tensor-owned loading to request an owned resident buffer from file-backed input.
• Handle copy-based allocation / transfer success and failure semantics explicitly.
• Integrate with the tensor-to-io boundary introduced by Add emel/io module and tensor-to-io orchestration boundary #60.

Hard Constraints

• This issue implements the read/copy strategy only.
• Tensor lifecycle/residency semantics remain owned by model/tensor.
• Do not fold mmap or staged/chunked constrained-memory policy into this issue.
• Keep the strategy boundary explicit enough that later cooperative loading can build on it cleanly.

Acceptance Criteria

• A dedicated read/copy strategy state machine exists under src/emel/io.
• Tensor-owned load flow can request copy-based residency through the io boundary.
• Strategy-specific errors are surfaced explicitly and deterministically.
• Tests cover success and representative transfer/allocation failures.

Related

• Depends on Add emel/io module and tensor-to-io orchestration boundary #60.