Fix dependency-name computation in schema-apply ordering

[aersam]

Two real bugs in how _iter_sql_files/_get_sql_deps compute the name a file is tracked under, both silently degrading dependency ordering to "untracked -> assumed already satisfied" instead of an actual wait:

• _get_sql_deps used str(t) on a matched Table node, which includes the alias for an aliased reference (e.g. "FROM app.visit v" -> "app.visit AS v"). That never matches the plain name a dependency's own CREATE target is recorded under. Use exp.table_name(t) instead.
• _iter_sql_files derived a file's own declared name from the raw file stem, so a numeric ordering prefix (e.g. "0_visit_unified.sql") landed in all_declared as "schema.0_visit_unified" instead of "schema.visit_unified" -- never matching what any dependent file's SQL actually references. Strip it the same way the schema-directory name already is.

apply_schema()'s retry-until-no-progress loop (this version) still converges even with these bugs present, since it retries raw SQL execution directly rather than relying on this bookkeeping - but the computed ordering itself was wrong, causing avoidable extra retry passes.