fix(ingest/teradata): correct nullable and autoincrement hydration for CHAR(N)-padded values

[JohnRTurner]

Summary

teradatasqlalchemy._get_column_info uses strict equality / membership checks that don't tolerate CHAR(N) space-padding:

• 'nullable': row['Nullable'] == 'Y' — Teradata returns 'Y ' / 'N ' (CHAR(1) padded), so the check evaluates False for every column. Genuinely nullable columns hydrate as nullable: false everywhere.
• autoinc = row['IdColType'] in ('GA', 'GD') — Teradata returns 'GA ' / 'GD ' (CHAR(4) padded), so the check evaluates False for every identity column. Identity columns hydrate as autoincrement: false.

Both bugs reproduce on a stock Teradata install with teradatasqlalchemy==20.0.0.2, no special configuration. The nullable bug is what surfaced the report (customer ticket: every column appears non-nullable); the IdColType bug is a latent sibling found by auditing other CHAR field accesses in the same code path.

Fix

After self._get_column_info(row) in optimized_get_columns, re-derive nullable and autoincrement from the raw row fields with whitespace-stripping. Shared _read_padded_char_field helper handles SQLAlchemy Row attribute access and dict access. Both helpers are no-ops when the raw value isn't a string (e.g. NULL — most often a view column without QVCI, where the metadata is genuinely unknown and guessing either direction would be wrong).

Test plan

• Unit: new TestCharPaddingFixes class with 7 focused tests covering padded 'Y '/'N ', clean 'Y', dict-row vs Row-attr access, padded 'GA ' and 'GD ', and IdColType=None no-op.
• Full tests/unit/test_teradata_source.py: 112 passing.
• ./gradlew :metadata-ingestion:lintFix: clean.
• End-to-end against a real local Teradata: padded 'Y ' now hydrates as nullable=True; padded 'GA ' now hydrates as autoincrement=True.

What this does and doesn't cover

✅ Padded 'Y ' / 'N ' for Nullable (the primary customer-reported bug)
✅ Padded 'GA ' / 'GD ' for IdColType (autoincrement detection)
❌ View columns where Nullable is genuinely None (e.g., views without QVCI enabled on Teradata) — the metadata isn't available, and the helper correctly does not guess. Addressed customer-side via QVCI on Teradata.

Out of scope

Other padded CHAR fields the library handles internally (ColumnType, ColumnName, etc.) — these flow through normalize_name or _resolve_type in the library, which already strip. Audited and confirmed no user-visible breakage from those.

🤖 Generated with Claude Code

[github-actions]

Linear: ING-2769

Thanks for your contribution! We have created an internal ticket to track this PR. A member of the core DataHub team will be assigned to review it within the next few business days - you will get a follow-up comment once a reviewer is assigned.