[amplifier-core] Module loader reports misleading error when import fails deep in dependency chain

Bug Description

When a module fails to load due to a dependency error deep in the import chain, the loader in amplifier-core reports the wrong error message. Instead of surfacing the root cause, it reports the top-level import failure, making debugging nearly impossible.

Reproduction

The error occurs when provider-anthropic tries to load in a sub-agent session:

Failed to load module 'provider-anthropic': Module 'provider-anthropic' failed validation:
FAILED: 0/1 checks passed (1 errors, 0 warnings).
Errors: module_importable: Failed to import module:
  cannot import name 'PROVIDER_THROTTLE' from 'amplifier_core.events'

The reported cause: PROVIDER_THROTTLE doesn't exist in amplifier_core.events
The actual cause: pydantic_core._pydantic_core native extension can't load (Python version mismatch between system Python 3.14 and uv tool env Python 3.12)

The full import chain:

amplifier_core.events
  → amplifier_core.__init__
    → amplifier_core.coordinator
      → amplifier_core.hooks
        → amplifier_core.models
          → pydantic
            → pydantic_core._pydantic_core  ← ACTUAL FAILURE (ModuleNotFoundError)

Python collapses this into an ImportError on the top-level import (PROVIDER_THROTTLE), hiding the real problem 5 levels deep.

Impact

• The error message says a constant is missing from amplifier_core.events — but it's actually there in the source
• A developer investigating this would look at events.py, find PROVIDER_THROTTLE present, and be completely confused
• The actual fix (reinstall pydantic-core or fix Python version) is invisible from the error message
• When this happens during delegate(), the sub-agent fails with RuntimeError: No providers mounted — also misleading

Suggested Fix

In amplifier_core/loader.py's _validate_module method, catch the ImportError and include the full exception chain:

# Current (misleading):
except ImportError as e:
    errors.append(f"Failed to import module: {e}")

# Proposed (includes root cause):
except ImportError as e:
    import traceback
    chain = traceback.format_exception(e)
    root_cause = e
    while root_cause.__cause__:
        root_cause = root_cause.__cause__
    errors.append(
        f"Failed to import module: {e}"
        f"\n  Root cause: {type(root_cause).__name__}: {root_cause}"
    )

This would produce:

Failed to import module: cannot import name 'PROVIDER_THROTTLE' from 'amplifier_core.events'
  Root cause: ModuleNotFoundError: No module named 'pydantic_core._pydantic_core'

Secondary Issue

There may also be a related issue where sub-agent sessions spawned via delegate() resolve to the system Python (3.14) instead of the uv tool environment's Python (3.12), causing native extensions compiled for 3.12 (.cpython-312-darwin.so) to fail to load. This is what triggers the pydantic_core failure in the first place. The misleading error message makes this Python version mismatch very hard to diagnose.

Environment

• amplifier-core: 1.0.0
• provider-anthropic: 1.0.0
• pydantic: 2.12.5 / pydantic-core: 2.41.5
• uv tool env Python: 3.12.10
• System Python: 3.14.2
• macOS arm64 (26.3)

Start in amplifier_core/loader.py at _validate_module and reproduce the provider-anthropic failure during a sub-agent session. Trace how the ImportError is assembled through the dependency chain and verify that the reported error includes the underlying pydantic_core failure, while checking whether the delegate() Python-environment issue is separate.