Add planner-based analysis pipeline and optional /api/analyze planner flow

[rad1092]

Motivation

• Provide a lightweight planner that can parse a natural-language analysis question into structured controls (top N, sample N, threshold, region, compare periods) and produce an executable plan.
• Allow the web analysis endpoint to optionally run that plan and return a machine-readable execution result alongside the existing analysis payload.

Description

• Add bitnet_tools/planner.py defining AnalysisIntent and AnalysisPlan dataclasses and implementing parse_question_to_intent, build_plan, execute_plan, and execute_plan_from_csv_text which build and run a filter -> groupby -> agg -> rank -> sample -> export execution graph and return table/sample/meta with graceful fallback on errors.
• Integrate planner into bitnet_tools/web.py by importing parse_question_to_intent, build_plan, and execute_plan_from_csv_text and adding a use_planner boolean option on /api/analyze that appends planner.intent, planner.plan, and planner.execution to the response when enabled.
• Add tests/test_planner.py to validate parsing, plan-node structure, and fallback behavior for invalid nodes.
• Extend tests/test_web.py with an integration test verifying that /api/analyze returns the planner payload when use_planner is true.

Testing

• Ran the new and updated automated tests with pytest -q tests/test_planner.py tests/test_web.py and all tests passed (15 passed in 1.74s).
• Tests cover intent parsing, plan generation, executor fallback behavior, and the web API planner integration.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2a8dd7313c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Exclude period tokens from region inference

When a question asks for a before/after comparison without naming a region (for example containing before, after, 이전, or 이후), this fallback path treats the first such token as intent.region. That makes the filter node run against region_column with a non-region value, which can drop all rows and return empty planner results for valid compare-period queries.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Execute planner on normalized input used by analysis

This execution step always reads request_payload.normalized_csv_text, but /api/analyze still accepts legacy csv_text and build_analysis_payload_from_request promotes it when normalized text is empty. In that common compatibility path, analysis summary is built from real rows while planner execution runs on an empty CSV, producing inconsistent/empty planner output despite a successful analysis response.

Useful? React with 👍 / 👎.