feat: Add TuShare data collector (incremental update & resume support)

[JakobWong]

Description

This PR introduces a new data collector for TuShare (daily frequency) under qlib/scripts/data_collector/tushare/collector.py.
It provides a robust ETL pipeline similar to the Yahoo collector but tailored for TuShare's API and A-share market features.

Key features:

1. Incremental Update & Resume:
   • Supports "resume from breakpoint" by checking existing CSVs and only downloading data newer than the local max date.
   • update_data_to_bin dumps only newly added dates (using a temporary directory) to improve performance, instead of full redump.
2. Data Consistency:
   • Includes listed, delisted, and paused stocks (status L,D,P) to avoid survivorship bias.
   • Normalizes output to Qlib standard: date, open, high, low, close, volume, [amount], factor, symbol. amount is optional.
   • Explicitly handles duplicates and ensures monotonic dates.
3. Robustness:
   • Prefers TuShare's trade_cal for calendar acquisition, with a fallback to Qlib's default.
   • Enforces a baseline requirement for incremental updates (raises error instead of auto-downloading incompatible Yahoo sample data).

Motivation and Context

The existing collectors (Yahoo) are less stable for CN market data. Users often need a production-ready TuShare collector that supports large-scale historical fetch (with rate limits) and daily incremental updates without redownloading entire history. This implementation fills that gap with a structure consistent with Qlib's existing collectors.

How Has This Been Tested?

• Pass the test by running: pytest qlib/tests/test_all_pipeline.py under upper directory of qlib.
• If you are adding a new feature, test on your own test scripts.

Test Details:
Verified with local unit/integration tests (pytest tests/test_tushare_collector.py - Note: test file not included in this PR to keep it minimal, but logic verified):

1. Normalization: Validated against fixed CSV fixtures (ensuring correct column mapping, date parsing).
2. Incremental Logic: Verified update_data_to_bin correctly identifies the incremental window and creates temp storage.
3. Baseline Check: confirmed it raises RuntimeError if qlib_data_1d_dir is missing/invalid during update.

Screenshots of Test Results (if appropriate):

1. Pipeline test: (Skipped as strict environment required)
2. Your own tests:
   (All passed locally)

   tests/test_tushare_collector.py ..... [100%]
   5 passed in 2.43s
   

Types of changes

• Fix bugs
• Add new feature
• Update documentation

---

[SunsetWolf]

Hi, @JakobWong , First of all thank you for contributing the code, I see that the description of this pull request implements a lot of functionality. However, it doesn't seem to be clear how to use these features, and I was wondering if it would be possible to add documentation or a docstring to help people understand how to use them.

[JakobWong]

Hi, @JakobWong , First of all thank you for contributing the code, I see that the description of this pull request implements a lot of functionality. However, it doesn't seem to be clear how to use these features, and I was wondering if it would be possible to add documentation or a docstring to help people understand how to use them.

Hi @SunsetWolf, Thanks for reviewing! The data collector I added is so that users can collect data via the tushare api easily (currently supporting only CN data, daily).

I added documentation for the TuShare daily collector in qlib/scripts/data_collector/tushare/README.md, covering prerequisites (TUSHARE_TOKEN), a one-shot pipeline command, step-by-step download/normalize/dump, incremental updates, and validation. I also listed the TuShare collector in the data_collector overview.

Please let me know if you’d like further details or more examples.