Add improved typed casting to BigQuery

[betodealmeida]

CATEGORY

Choose one

• Bug Fix
• Enhancement (new features, refinement)
• Refactor
• Add tests
• Build / Development Environment
• Documentation

SUMMARY

Add improved type casting for BigQuery, similar to what #8226 did for Presto. This prevents 64 bit integers from being converted to floats by Pandas, losing precision.

TEST PLAN

I verified that int64 values are not cast to float, and that precision is maintained. I also tested that all types are cast correctly:

SELECT
  'string' AS `string`,
  CAST('string' AS BYTES) AS `bytes`,
  100 AS `integer`,
  100.0 AS `float`,
  CAST(1 AS NUMERIC) AS `numeric`,
  CAST(1 AS BOOLEAN) AS `boolean`,
  STRUCT(1) AS `record`,
  TIMESTAMP(DATETIME "2008-12-25 15:30:00", "America/Los_Angeles") AS `timestamp`,
  DATE "2008-12-25" AS `date`,
  TIME "15:30:00" AS `time`,
  DATETIME "2008-12-25 15:30:00" AS `datetime`

Generated CSV:

string	bytes	integer	float	numeric	boolean	record	timestamp	date	time	datetime
string	b'string'	100	100.0	1.0	True	"{""_field_1"": 1}"	2008-12-25 23:30:00+00:00	2008-12-25	15:30:00	2008-12-25 15:30:00

ADDITIONAL INFORMATION

• Has associated issue:
• Changes UI
• Requires DB Migration.
• Confirm DB Migration upgrade and downgrade tested.
• Introduces new feature or API
• Removes existing feature or API

REVIEWERS

[mistercrunch]

LGTM

[robdiciuccio]

Hey Beto, I've been working on this problem a bit for Postgres and MySQL, and I'm currently testing casting ints as object (strings) rather than the new Int64 type, since the latter is incompatible with PyArrow, which has significant serialization performance benefits. At the same time, it may go towards resolving the JS max integer size issue. Happy to chat with you about this.

[betodealmeida]

Will merge, but work on a better solution with @robdiciuccio.