fix: move LLMError into the llm package (#1096)

[CoronRing]

Summary

• Moves LLMError from railtracks.exceptions into railtracks.llm.errors, inheriting from RTLLMError instead of RTError, so the llm package no longer imports from railtracks.exceptions
• Also removes a secondary violation: NodeInvocationError (a railtracks exception) was being raised inside llm/models/_litellm_wrapper.py; replaced with ModelError from within the llm package
• Updates all consumers (built_nodes, tests) to import LLMError from railtracks.llm or railtracks.llm.errors
• Exports LLMError from railtracks.llm.__init__ for discoverability

Files changed

File	Change
llm/errors.py	New — LLMError(RTLLMError) definition
llm/__init__.py	Export LLMError
llm/models/_litellm_wrapper.py	Import LLMError from ..errors; replace NodeInvocationError with ModelError
exceptions/errors.py	Remove LLMError class
exceptions/__init__.py	Remove LLMError export
6× built_nodes/concrete/*.py	Update imports
3× test files	Update imports; update NodeInvocationError → ModelError test expectation

Test plan

• test_litellm_wrapper.py — all tests pass, including the updated ModelError expectation for invalid tool parameters
• test_structured_tool_call_llm.py — all tests pass
• test_exceptions.py — LLMError tests pass with new location
• Full unit suite: 1508 passed

Closes #1096