[TEMP PR] feat(ByteDance): new Layer Separation node

[bigcat88]

WIP

API Node PR Checklist

Scope

• Is API Node Change

Pricing & Billing

• Need pricing update
• No pricing update

If Need pricing update:

• Metronome rate cards updated
• Auto‑billing tests updated and passing

QA

• QA done
• QA not required

Comms

• Informed Kosinkadink

[christian-byrne]

Closing as superseded — verified that nothing is stranded.

Successor: #15351 (the partner node, standalone, approved)
Also: #15317 (the compositor half)

This branch was stacked: it carried Terry's two compositor commits verbatim plus four of Alexander's own. Both halves are fully preserved elsewhere:

• Alexander's 319 lines (comfy_api_nodes/nodes_bytedance.py +302, comfy_api_nodes/apis/bytedance.py +17) are byte-identical in [Partner Nodes] feat(ByteDance): Seedream Layer Separation node #15351. git diff 8d3300c9:<file> 4931337c:<file> is empty for both. Alexander's four commits here (73c5baa5, 5243a245, d90dc467, 8d3300c9) touched only those two files, so nothing of his lives in the other four.
• Terry's four files (comfy_api/latest/_io.py, nodes.py, comfy_extras/nodes_compositor.py, comfy_extras/compositor_blend.py) are byte-identical to commit 19467819 here, and 19467819 is an ancestor of [Partner Nodes] feat: ImageCompositor node with layer-state compositing, layer from bbox and Seedream Layer Separation node #15317 head 9232943a. That PR is three commits further along (b6b2572c review feedback, 9e5cb779 blend-mode boundary alignment, 9232943a single batched inputs), so keeping this branch alive would only risk a hand-resolved merge reverting them.

No reviews, no review comments and no issue comments on this PR, so no discussion is lost either.

Closing this also removes the squash-merge hazard: with both PRs open, a squash landing of #15317 would have made this branch conflict on three shared files, with a real chance of a resolution that reverted the MAX_RESOLUTION canvas guard added in b6b2572c.