Add call-and-response rework workflow, streamline README, and deprecate legacy wrappers

[velocitylord]

Motivation

• Provide a presentation-ready generation workflow that produces more musical outputs by combining an intact original phrase with a predictor-generated response.
• Reduce repository clutter and focus documentation on the core, demo-relevant pipelines while preserving the key/chord analysis functionality.

Description

• Add Scripts/callAndResponseRework.wls, a Wolfram script that implements a bar-structured flow where bars 1–2 keep the original melody and bars 3–4 are generated as a harmony-aware predictor response, with options for responseKeyShift and rhythmBlend, and export of MIDI + analysis JSON (global key, bar timeline, call/response spans, generation settings).
• Deprecate older tooling by replacing Scripts/keyShiftAndRegenerateMelody.wls with a one-line deprecation stub and turning Scripts/rework_with_prettymidi.py into a short-exit wrapper pointing users to the new scripts.
• Streamline README.md to highlight the core demo workflow (makeDataset, trainRnn, reinterpretPhraseWithHarmony, and the new callAndResponseRework) and provide concise example invocations for the presentation path.
• Small cleanups and conservative defaults maintained/refined in Scripts/reinterpretPhraseWithHarmony.wls (argument validation, harmony/key detection, conservative maxPitchDelta/keepOriginalProb/timingBlend behavior, and stable analysis export).

Testing

• Ran python -m py_compile Scripts/rework_with_prettymidi.py which succeeded.
• Verified repository changes and references with a grep scan rg -n "keyShiftAndRegenerateMelody|rework_with_prettymidi|callAndResponseRework|reinterpretPhraseWithHarmony" README.md Scripts which succeeded.
• Attempted to validate Wolfram script behavior with wolframscript but the wolframscript binary is not available in this environment, so runtime integration tests for .wls scripts could not be executed here.

---

Codex Task

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 343cb39dc8

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]

Avoid double-transposing fallback response pitches

In the predictor-failure path, predPitch is set to anchorPitch, and anchorPitch already has responseKeyShift applied. The later call chooseConstrainedPitch[predPitch + responseKeyShift, ...] therefore applies the key shift a second time whenever responseKeyShift != 0, so failed predictions produce systematically over-shifted notes instead of the intended transposition. This is most visible when the model occasionally returns non-numeric output or throws, because the fallback melody drifts sharper/flatter than requested.

Useful? React with 👍 / 👎.