Prefer input columns over GROUP BY aliases

[osipovartem]

Which issue does this PR close?

• Follow-up for Snowflake/PostgreSQL-compatible GROUP BY alias precedence on the embucket-sync-df55.0.0 branch.

Rationale for this change

When a bare GROUP BY identifier matches both an input column and a SELECT-list alias, SQL resolves it to the input column. DataFusion assembled the input and projected schemas before parsing the expression, so a qualified input field plus an unqualified alias with the same name produced a false ambiguity error.

What changes are included in this PR?

• Resolve a bare GROUP BY identifier against the input schema first when that input column is unambiguous.
• Keep the existing projected-schema path for alias-only names and non-bare expressions.
• Use the selected schema consistently for normalization and placeholder inference.

Are these changes tested?

Yes. Added planner tests for both directions:

• SELECT SUM(id) AS age FROM person GROUP BY age groups by the input person.age.
• SELECT state AS age FROM person GROUP BY age reaches the expected non-aggregate validation instead of a false ambiguity.

Local verification:

cargo test -p datafusion-sql --test sql_integration
584 passed

cargo clippy -p datafusion-sql --all-targets --all-features -- -D warnings
passed

Are there any user-facing changes?

Yes. Queries with a SELECT alias that collides with an input column now follow SQL GROUP BY input-column precedence. There is no public API change.