feat(orchestrator): Harden VFS and enhance artifact awareness across workers

backend/src/ea_chatbot/graph/prompts/coder.py

@@ -9,6 +9,14 @@ The user will provide a task and a plan.
 - Do NOT assume a dataframe `df` is already loaded unless explicitly stated. You usually need to query it first.
 - The database schema is described in the prompt. Use it to construct valid SQL queries.
 
+**Virtual File System (VFS):**
+- An in-memory file system is available as `vfs`. Use it to persist intermediate data or large artifacts.
+- `vfs.write(filename, content, metadata=None)`: Save a file (content can be any serializable object).
+- `vfs.read(filename) -> (content, metadata)`: Read a file.
+- `vfs.list() -> list[str]`: List all files.
+- `vfs.delete(filename)`: Delete a file.
+- Prefer using VFS for intermediate DataFrames or complex data structures instead of printing everything.
+
 **Plotting:**
 - If you need to plot any data, use the `plots` list to store the figures.
 - Example: `plots.append(fig)` or `plots.append(plt.gcf())`.
@@ -18,7 +26,8 @@ The user will provide a task and a plan.
 - Produce FULL, COMPLETE CODE that includes all steps and solves the task!
 - Always include the import statements at the top of the code (e.g., `import pandas as pd`, `import matplotlib.pyplot as plt`).
 - Always include print statements to output the results of your code.
-- Use `db.query_df("SELECT ...")` to get data."""
+- Use `db.query_df("SELECT ...")` to get data.
+"""
 
 CODE_GENERATOR_USER = """TASK:
 {question}
@@ -43,6 +52,7 @@ Return a complete, corrected python code that incorporates the fixes for the err
 - You have access to a database client via the variable `db`.
 - Use `db.query_df(sql)` to run queries.
 - Use `plots.append(fig)` for plots.
+- You have access to `vfs` for persistent in-memory storage.
 - Always include imports and print statements."""
 
 ERROR_CORRECTOR_USER = """FAILED CODE: