Feat(snowflake): improve transpilation of unnested object lookup

[georgesittas]

This PR:

• Introduces a new transformation for Snowflake, in order to transpile BigQuery queries with unnests to it
• Modifies the walk_in_scope logic to stop considering plain UDTFs without Query children as scope boundaries

We incorrectly transpile BigQuery queries like the following to Snowflake today:

import sqlglot
print(sqlglot.transpile("select _u.foo from unnest([struct('x' as foo, 'y' AS bar)]) as _u", "bigquery", "snowflake")[0])
# SELECT _u.foo FROM TABLE(FLATTEN(INPUT => [OBJECT_CONSTRUCT('foo', 'x', 'bar', 'y')])) AS _t0(seq, key, path, index, _u, this)

The resulting query fails with Error: invalid identifier '_U.FOO' (line 1), because the value _u is a VARIANT, and so accessing its contents using the dot notation is not allowed. Instead, Snowflake requires that we use either the colon (x:y) or the bracket syntax x['y'] for this reason.

To address this, the new transformation looks for unnests that have a single column alias, and then replaces any columns in scope that are qualified with said alias.

I looked into Snowflake's docs for TABLE(FLATTEN(...)), and I believe the expected receivers of the lookup operation should be OBJECT or VARIANT values, for which it is safe to replace the dot syntax with the bracket notation.