Fix column type OIDs in pgduck_server row description metadata

[Khoulii]

Description

Fixes #67

Currently, duckdb_query_result_send_column_metadata always writes InvalidOid (0) as the column type OID in the row description message, regardless of the actual column type. This breaks clients that rely on PQftype() for type introspection (e.g. ORMs, psycopg2, JDBC drivers).
This PR adds duckdb_type_to_pg_oid() in type_conversion.c which maps duckdb_type values to their closest PostgreSQL OID equivalents, and wires it into the column metadata sender in place of the hardcoded InvalidOid.

Mapping strategy:

• Exact matches where sizes align: BOOLEAN→BOOLOID, SMALLINT→INT2OID, INTEGER→INT4OID, BIGINT→INT8OID, FLOAT→FLOAT4OID, DOUBLE→FLOAT8OID, DATE→DATEOID, TIME→TIMEOID, TIMESTAMP→TIMESTAMPOID, TIMESTAMPTZ→TIMESTAMPTZOID, INTERVAL→INTERVALOID, UUID→UUIDOID, BLOB→BYTEAOID
• Unsigned types promoted to the next signed PG size that fits (UTINYINT→INT2OID, USMALLINT→INT4OID, UINTEGER→INT8OID)
• Types with no native PG equivalent (UBIGINT, HUGEINT, UHUGEINT, DECIMAL) → NUMERICOID
• Complex/nested types (LIST, ARRAY, STRUCT, MAP, UNION, ENUM) → TEXTOID since they are serialised as text. These types should be considered to be handled more properly in future PRs.

columnLength and columnTypeMod remain -1; NUMERIC typmod encoding (precision/scale) is left as a follow-up.

---

Checklist

• I have tested my changes and added tests if necessary
• I updated documentation if needed
• I confirm that all my commits are signed off (DCO)

[sfc-gh-mslot]

Thanks for looking into this! Did you run into a specific issue with the invalid OIDs?

[Khoulii]

Thanks for looking into this! Did you run into a specific issue with the invalid OIDs?

@sfc-gh-mslot
thank you for the follow up,
the current behavior is to always return invalid OIDs, which is not a good thing to have.
I acknowledge that in production environment this won't be an issue as pgduck_server is not handling the requests, as it will be processed by the actual pg server.
but for local/dev and some CI/CD it's good to have a well-behaving duckdb over postgres wire protocol without bearing the burden of a heavy weight pg server.

this issue reports that Grafana was unable to draw graphs ..etc. because of the bad OIDs.

again, we did not observe this on production but it's very convenient for local/dev use cases.

[sfc-gh-dachristensen]

Triggering PR vaildation now. Thanks for the contribution!

[sfc-gh-mslot]

@sfc-gh-dachristensen this requires extensive test changes

I think I got most of them at
https://github.com/Snowflake-Labs/pg_lake/tree/marcoslot/fix-column-oids-327

[sfc-gh-dachristensen]

@sfc-gh-mslot branch looks good; we want to force push here?