fix(freshcore): fall back for non-scalar dtypes and wide ints, restore integer dtypes, keep non-string labels

[kevincostner17]

Summary

FreshCore's native arrays only carry float, bool and string columns, and the adapter rebuilt results from plain lists keyed by str(label). That silently changed dtypes and labels. This PR:

• Falls back for datetime (naive and tz-aware), timedelta, categorical, period and interval columns, which used to come back as strings.
• Falls back for integer columns holding a value beyond ±2**53, which float64 cannot represent exactly.
• Restores integer dtypes (int64, nullable Int64, other widths) when every returned value is integral and in range. Missing values are allowed only for nullable dtypes. Otherwise the column stays float64 and the change is recorded in report.backend_differences (step "dtypes").
• Falls back when distinct column labels collide once stringified, before or after renaming (1 and "1", 1 and " 1 "). Previously one column silently overwrote the other.
• Maps non-colliding non-string labels (e.g. 0, 1 from a headerless CSV) back to the original labels.

All new fallbacks use the existing fallback path, so fallback_policy="error" raises FallbackError. docs/fallback-matrix.md, docs/backends.md and docs/freshcore.md list the new triggers.

Not changed here: bool columns still come back as nullable boolean, and float32 / nullable Float64 still come back as float64.

Tests

New tests/test_execution/test_freshcore_dtype_fidelity.py uses a fake native module:

• Each fallback (datetime, tz-aware datetime, categorical, timedelta, period, interval, several kinds at once, wide Int64/int64/uint64, colliding labels before and after renaming) gives pandas-equal output, dtypes and actions, with the recorded reason. Each also raises FallbackError under fallback_policy="error".
• The FreshCore engine returns datetime/category columns as strings and ints as float64 #262 reproduction frame matches the pandas engine's dtypes, and big[0] == 2**53 + 1.
• int64, Int64, Int8 and uint16 round-trip through the native path with their dtype and match the pandas engine.
• Values that are non-integral, missing from a non-nullable int column, or out of range stay float64 and are recorded in backend_differences.
• Integer, RangeIndex and mixed renamed/non-string labels round-trip and match the pandas engine.
• Supported shapes (ints at exactly ±2**53, all-NA Int64, distinct mixed labels) stay native.

Verification

• ruff check .: passes
• mypy src/freshdata: no issues (202 files)
• Python 3.12: 5078 passed, 6 skipped
• Python 3.9 / pandas 1.5.3: 5074 passed, 10 skipped
• Merges cleanly with main (includes fix(freshcore): fall back for missforest, per-column strategies, non-finite outliers and nullable bools; honour native duplicate counts #375)

Closes #262
Refs #232 (part 2)

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: c1a9fe3c-72f5-4af4-a967-0ffa89c4419b

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

FreshData benchmark report — performance

• freshdata: ?
• python: ?
• platform: ?

fixture	n_rows	n_cols	p50 s	p95 s	peak MB	repair %	false-repair %	preserve %	trust	monotonic	export %

Authored-code reduction (Metric 6)