feat: per-call model/provider override on delegate_task

[HenkDz]

Closes #3719

Summary

Add model and provider parameters to delegate_task (top-level and per-task in batch mode) so the parent agent can dynamically route subagents to different models based on task type.

Motivation

Currently subagent model selection is static via delegation.model/delegation.provider in config.yaml — all subagents use the same model. Different tasks benefit from different models (code → fast model, research → strong reasoning model).

Changes

tools/delegate_tool.py (136 insertions, 34 deletions)

1. Schema: Added model and provider as top-level params + inside batch task items
2. _resolve_call_credentials(): New helper that merges per-call overrides with config-level credentials via resolve_runtime_provider()
3. delegate_task(): Refactored to resolve credentials per-task instead of once globally. Top-level model/provider flow into single-task mode; batch mode uses each task's own overrides
4. Registry handler: Wires model/provider through the lambda

Resolution Priority

per-call model/provider  →  config delegation.model/provider  →  inherit parent

Fully backward compatible — omitting model/provider preserves existing behavior.

Example Usage

Single task:

delegate_task({
  "goal": "Implement auth middleware",
  "model": "glm-5.1",
  "provider": "zai"
})

Batch with per-task routing:

delegate_task({
  "tasks": [
    {"goal": "Research competitors", "model": "glm-5", "provider": "zai"},
    {"goal": "Write integration tests", "model": "glm-5.1"}
  ]
})