feat: add support for env_vars to ReasoningEngine.create

[EaCognitive]

Problem

The ReasoningEngine.create method in the Vertex AI SDK currently lacks support for passing environment variables (env_vars) to the reasoning engine at creation time. This is a fundamental requirement for many LangChain-based agents that rely on environment variables for configuration and secrets. While the newer AgentEngine class supports this, ReasoningEngine has fallen behind in feature parity.

Solution

This PR adds support for env_vars to ReasoningEngine.create. It also refactors the shared deployment specification logic into a new _utils.py module to ensure consistent behavior across both AgentEngine and ReasoningEngine.

Changes

• Created vertexai/reasoning_engines/_utils.py to host shared deployment specification helper functions.
• Updated vertexai/reasoning_engines/_reasoning_engines.py to accept and pass the env_vars parameter during resource creation.
• Aligned ReasoningEngine with the AgentEngine implementation pattern.

Verification

• Verified that env_vars are correctly mapped into the ReasoningEngineSpec.DeploymentSpec.
• Consistent with established SDK patterns.

[EaCognitive]

Updated with current main and re-ran targeted coverage. Local validation: uv run ... pytest tests/unit/vertex_langchain/test_reasoning_engines.py -k 'test_create_reasoning_engine_with_env_vars or test_generate_deployment_spec' -> 6 passed. The branch preserves existing Reasoning Engine utilities and adds env var name, direct value, SecretRef, invalid type, and missing env-var coverage.