[SPARK-57457][SQL] Support nanosecond-precision timestamp types in the CSV datasource (v1 and v2)

[vinodkc]

What changes were proposed in this pull request?

This PR adds nanosecond-precision timestamp support (TIMESTAMP_NTZ(p) and TIMESTAMP_LTZ(p)) to the CSV datasource, for both the v1 (CSVFileFormat) and v2 (CSVTable) paths.

Specifically:

• Parser (UnivocityParser): adds TimestampNTZNanosType and TimestampLTZNanosType cases that delegate to the existing parseWithoutTimeZoneNanos / parseNanos formatter methods.
• Generator (UnivocityGenerator): adds the corresponding write-path cases that delegate to formatWithoutTimeZoneNanos / formatNanos.

Why are the changes needed?

CSV rejected nanos timestamp types in its datasource capability checks and lacked the conversions to round-trip them, so these columns could not be written or read through CSV.

Does this PR introduce any user-facing change?

Yes. Users can write and read TimestampNTZNanosType(p) / TimestampLTZNanosType(p) (p in 7..9) with CSV

How was this patch tested?

• CsvFunctionsSuite — updated the existing from_csv nanosecond timestamp test: the test now asserts successful parsing and correct truncated value rather than expecting an UNSUPPORTED_DATATYPE exception.

• FileBasedDataSourceSuite — new end-to-end round-trip test covering both v1 and v2 source paths, precisions (7–9), and both TimestampNTZNanosType and TimestampLTZNanosType, verifying that a DataFrame written to CSV and read back with a matching schema produces identical results.

Was this patch authored or co-authored using generative AI tooling?

Yes, Generated-by: Claude Code (Sonnet 4.6) was used to assist with this patch.

[MaxGekk]

0 blocking, 0 non-blocking, 0 nits. LGTM — a minimal, faithful extension of the CSV datasource to nanosecond TIMESTAMP_NTZ(p) / TIMESTAMP_LTZ(p) (p ∈ 7..9), mirroring the micro-precision converters and dropping the capability gate on both v1 (CSVFileFormat) and v2 (CSVTable).

Verification

The read/write paths are symmetric (parseWithoutTimeZoneNanos/parseNanos ↔ formatWithoutTimeZoneNanos/formatNanos), all four formatter methods exist with matching signatures over TimestampNanosVal, and both capability gates are updated (the prior "not supported" test correctly drops CSV). Two deliberate behaviors I checked and cleared: the nanos LTZ case omits the micro TimestampType legacy-parse fallback (appropriate — nanos has no Spark 1.x/2.0 legacy data, and micro TimestampNTZType has none either), and the 3-digit default timestamp format truncates sub-millisecond digits for all timestamp types (pre-existing, not a nanos regression; tests use explicit formats). Coverage is solid: FileBasedDataSourceSuite round-trips v1+v2 × NTZ/LTZ × precisions 7–9 (write → read-back → checkAnswer), and the updated from_csv test validates the read path against an independently-authored string with precision truncation.

@HyukjinKwon — could you take a look as well, since you've been close to the CSV/Univocity datasource paths?

[MaxGekk]

+1, LGTM. Merging to master/4.x.
Thank you, @vinodkc and @dongjoon-hyun @HyukjinKwon for review.