fix: keep manual bucket when autotuning from Python

[bact]

• Don't zero a user-set bucket when autotune is on
• Autotune's n-gram trials used to hash modulo 0 and crash
• A manual bucket=0 means no n-grams, so don't sample them

[Copilot]

Copilot review overview

🟡 Changes recommended

Manual zero buckets can still crash autotuning, and the regression test may not reliably exercise the failing path.

Review effort: Lite
Findings: 1

Open (1)

• Reject zero manual buckets before autotune dictionary hashing · New

What changed in this PR

This PR preserves manually configured buckets during Python autotuning and adds regression coverage. Non-positive buckets can still cause modulo-by-zero crashes, and the test depends on a timing-sensitive trial.

Changes:

• Retains user-specified buckets during autotuning.
• Adds regression coverage for manual bucket autotuning.
• Requires validation or substitution for non-positive buckets and deterministic trial coverage.

File	Summary
python/​fasttext_module/​fasttext/​FastText.py	Preserves manual bucket configuration during autotuning.
python/​fasttext_module/​fasttext/​tests/​test_autotune_bucket.py	Adds regression coverage for autotune bucket handling.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[bact]

Fixed as suggested in review. A manual bucket=0 under autotune now means no n-gram trials (no wordNgrams/minn sampling, maxn=0).