fix(e6): alias unaliased Databricks colon projections by final path segment

[tkaunlaky-e6]

Problem

A Databricks colon/variant projection written without an alias — e.g. SELECT deep_dives:metrics_per_level — is implicitly named by Databricks after its final path segment (metrics_per_level). E6 has no such naming rule, so when an outer query references that column by name it fails:

Column 'metrics_per_level' not found in any table

This is the "colon path stripped by transpiler" failure bucket.

Fix

For Databricks-sourced transpiles (from_dialect="databricks"), make the implicit name explicit: a bare colon/variant JSONExtract appearing as a top-level SELECT projection is wrapped in AS <final-segment>.

Scoped precisely by iterating the Select's own expressions, so these are left untouched:

• colon access inside WHERE / GROUP BY / a function / CASE (not a top-level projection)
• an already-aliased projection (… AS x)
• a path ending in a subscript (col:arr[0] — no implicit key name)
• Snowflake variant access (databricks-gated)

Example

Databricks: SELECT price FROM (SELECT box:price FROM orders) t
e6 (before): SELECT price FROM (SELECT box:price FROM orders) AS t -- inner col unnamed -> "not found"
e6 (after): SELECT price FROM (SELECT box:price AS price FROM orders) AS t -- resolves

Validation

• Live E6 engine: the aliased form resolves the outer reference and returns rows; the unaliased form fails with Column not found. Edge cases confirmed on the engine — reserved-word segment (box:limit → box:"limit" AS "limit") works; a duplicate final segment (a:x, b:x) reports ambiguous, which is faithful (Databricks also names both columns x).
• tests/dialects/test_e6.py: existing top-level colon projections now carry their implicit alias, plus a new subquery-reference regression test. Full suite: 59 passed / 900 subtests.