fix(databricks)!: transpile TRY_DIVIDE to/from other dialects [CLAUDE]

[baruchoxman]

Summary

Databricks' TRY_DIVIDE(a, b) returns NULL when the divisor is 0 (instead of raising an error). Previously it was left as exp.Anonymous, so it passed through untranspiled.

TRY_DIVIDE also exists in Spark (https://spark.apache.org/docs/latest/api/sql/index.html#try_divide), so this fix places the parsing and generation logic in the Spark dialect — Databricks inherits it automatically.

The fix maps TRY_DIVIDE to the existing exp.SafeDivide, which already encodes the exact same semantics. The base generator's safedivide_sql then handles transpilation to all target dialects correctly.

Changes:

• parsers/spark.py: parse TRY_DIVIDE(a, b) → exp.SafeDivide.from_arg_list (alongside TRY_ADD, TRY_MULTIPLY, TRY_SUBTRACT)
• generators/spark.py: emit native TRY_DIVIDE when targeting Spark/Databricks

Transpilation examples:

Source	Target	Output
Spark / Databricks	Spark / Databricks	TRY_DIVIDE(a, b)
Spark / Databricks	Snowflake	IFF(b <> 0, a / b, NULL)
Spark / Databricks	DuckDB	CASE WHEN b <> 0 THEN a / b ELSE NULL END

Fixes #7312

Test plan

• test_try_divide in tests/dialects/test_spark.py — verifies parsing from both spark and databricks and cross-dialect transpilation
• Full Spark and Databricks test suites pass

[baruchoxman]

@geooo109 handled the comments, please take another look

[georgesittas]

Thanks for the contribution, @baruchoxman.