feat: start v0.7.5 guardrails

[wxai-space]

Summary

• bump development version to 0.7.5
• add opt-in input, tool-call, and non-streaming output guardrails
• expose GuardrailDecision and GuardrailManager
• add stable LA-GUARDRAIL error code and guardrail_block trace events
• add guardrails docs, FAQ entry, and focused regression tests

Compatibility

• guardrails are disabled by default
• default agent.run() and stream=True behavior remain unchanged unless guardrails are configured
• output guardrails apply to non-streaming final responses in this initial implementation

Tests

• /Users/macpro/opt/anaconda3/envs/Langchain-Chatchat/bin/python -m compileall -q LightAgent tests/test_v065_core.py tests/test_v070_tracing.py tests/test_memory_policy.py tests/test_guardrails.py
• PYTHONPATH=. /Users/macpro/opt/anaconda3/envs/Langchain-Chatchat/bin/python -m pytest -q tests/test_v065_core.py tests/test_v070_tracing.py tests/test_memory_policy.py tests/test_guardrails.py