[e2e][python] Stabilize standalone nightly example tests

[wenjin272]

Linked issue: #642

Purpose of change

This is a follow-up to #708 that makes the standalone nightly example test reliable enough to validate example execution instead of only successful submission.

The change:

• builds and installs Flink Agents from the checked-out source and defaults the nightly run to Flink 2.3.0;
• waits for submitted jobs to become stably running or finish, propagates discovery/setup/cleanup failures, and adds focused Bats coverage for the shell helpers;
• shares the RAG Chroma data across the submitting process and TaskManager process;
• synchronizes lazy Chroma client initialization to avoid the RustBindingsAPI initialization race;
• uses a longer Ollama request timeout and disables thinking for the RAG example to keep CPU-only CI inference reliable; and
• ignores repository-local .codex configuration.

Tests

• uv run --no-sync pytest flink_agents/integrations/vector_stores/chroma/tests/test_chroma_vector_store.py flink_agents/examples/rag/tests/test_rag_agent.py -q — 23 passed, 1 skipped
• bats tools/test/unit/verify_example_job.bats — 25 passed
• Ruff check and format verification for the modified Python files
• Nightly standalone E2E with Flink 2.3.0: 13 passed, 0 failed — https://github.com/wenjin272/flink-agents/actions/runs/31109016948

API

No public API changes.

Documentation

• doc-needed
• doc-not-needed
• doc-included

Was this patch authored or co-authored using generative AI tooling?

• Yes
• No

Generated-by: Codex CLI 0.144.5