fix(bigquery)!: Do not normalize JSON fields in dot notation

[VaggelisD]

In BigQuery, JSON field lookups in dot notation are case sensitive, e.g:

bq> WITH t AS (SELECT PARSE_JSON('{"fOo": {"BaR": 1}}') AS bla) SELECT t.bla.fOo.BaR FROM t;
BaR
1

bq> WITH t AS (SELECT PARSE_JSON('{"fOo": {"BaR": 1}}') AS bla) SELECT t.bla.foo.bar FROM t;
bar
NULL

However, up until this point SQLGlot normalized all BigQuery identifiers as case insensitive which is valid for STRUCT lookups or other identifiers but may alter the semantics of these JSON accesses.

This PR fixes this behavior through the following steps:

1. During normalize_identifiers: Preserve the original column+dot parts in the column's meta
2. During qualify_columns: Distinguish the column from the dot parts i.e keeping only the latter in the meta
3. During annotate_types: Once all of the types are fully known, traverse JSON columns upwards to revert/repair the dot parts

[VaggelisD]

Here's my research thus far:

Case sensitivity

Dialect	JSON Case sensitivity
BigQuery	✅
Snowflake	✅
Databricks	✅
DuckDB	✅
Clickhouse	✅

Repros:

snowflake> WITH t AS (SELECT PARSE_JSON('{"a": {"A": 1}, "A": {"a": 2}}') AS col) SELECT col:a.A, col:A.a from t;
COL:A.A | COL:A.A
-- | --
1 | 2

databricks> WITH t AS (SELECT PARSE_JSON('{"a": {"A": 1}, "A": {"a": 2}}') AS col) SELECT col:a.A, col:A.a from t;
A	a
1	2

duckdb> WITH t AS (SELECT JSON '{"a": {"A": 1}, "A": {"a": 2}}' AS col) SELECT col.a.A, col.A.a from t;
┌──────┬──────┐
│  A   │  a   │
│ json │ json │
├──────┼──────┤
│ 1    │ 2    │
└──────┴──────┘

clickhouse (online playground)> WITH t AS (SELECT '{"a": {"A": 1}, "A": {"a": 2}}'::JSON AS col) SELECT col.a.A, col.A.a from t;
col.a.A | col.A.a 
    1        2        

Other dialects

• Postgres: Does not have dot notation, JSON extract operator requires string literals for names:

postgres> WITH t AS (SELECT '{"a": {"A": 1}, "A": {"a": 2}}'::JSON AS col) SELECT col.a.A, col.A.a from t;
ERROR:  missing FROM-clause entry for table "a"

postgres> WITH t AS (SELECT '{"a": {"A": 1}, "A": {"a": 2}}'::JSON AS col) SELECT col->a->A, col->A->a from t;
ERROR:  column "a" does not exist
                                                             ^
postgres> WITH t AS (SELECT '{"a": {"A": 1}, "A": {"a": 2}}'::JSON AS col) SELECT col->'a'->'A', col->'A'->'a' from t;
 ?column? | ?column?
----------+----------
 1        | 2
(1 row)

• Presto/Trino: Dot notation is limited to ROW objects:

trino> WITH t AS (SELECT CAST('{"a": {"A": 1}, "A": {"a": 2}}' AS JSON) AS col) SELECT col.a.A, col.A.a from t;
Query 20251114_094456_00003_maqp5 failed: line 1:81: Expression col is not of type ROW

[georgesittas]

Nice work @VaggelisD, seems like Toby's hunch was right. Let's get rid of the flag and just assume that JSON values will always have case-sensitive keys.