Add block weight text sync and chaining

[diodiogod]

This adds a block weight text field/output for the selective LoRA loaders and V2 analyzer+loader nodes.

What it does:

• adds a block_weights_string input
• adds a weights_output string output
• keeps the text field in sync with slider/toggle changes
• lets a positional weight string drive the UI back the other way
• allows chaining block profiles between nodes

Why:

• presets are useful, but they are manual UI state
• a text profile is easier to pass between nodes in larger workflows
• it makes block setups easier to save, reuse, tweak, and combine
• it is a more modular way to work with selective LoRA profiles

Notes:

• the generated output format is positional (1.00, 0.00, 0.50, ...)
• positional strings sync live with the UI
• named syntax is also supported on the Python side, using an Inspire Pack-style format adapted to these architectures
  • SDXL example: %default=1.0, te1=0.5, in7-8=1.2
  • FLUX example: %default=1.0, double=0.8, double7-12=1.2
  • Z-Image example: %default=1.0, layer18-25=1.2
• named syntax is not fully mirrored back into the UI live
• this is limited to the selective loader path and does not change the model layer editor preset system

[diodiogod]

Compatibility note for #56 / #57: when both branches are combined, connected Krea 2 block_weights_string values are ignored because Krea 2 is missing from the shared Python profile parser. The node falls back to its local slider values.

Tested fix: register the 28 Krea blocks, accept 28-value profiles or 29 values including other_weights, support named syntax, and include other_weights in the output. Verified with a connected runtime string on both Krea nodes.

Fix is preserved in diodiogod:personal/krea2-v1-save at commit 396970d. It should be applied when whichever PR merges second is rebased.