feat: add first and last aggregate functions

Substrait already defines `first_value` and `last_value`,
however they're defined as window functions.
First and last are commonly used as aggregate functions,
and having them as aggregates allows using them also in windows
(but not the other way around), see
https://substrait.io/expressions/window_functions/#aggregate-functions-as-window-functions

Spark defines first and last as aggregate functions
https://github.com/apache/spark/blob/a50b30d7a02bf45f1ddb8db0be6779b1441e1d4d/sql/core/src/main/scala/org/apache/spark/sql/functions.scala#L518
, and e.g. dropDuplicates is implemented as an aggregation using first()
https://github.com/apache/spark/blob/a50b30d7a02bf45f1ddb8db0be6779b1441e1d4d/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala#L2262

extensions/functions_arithmetic.yaml

@@ -1563,6 +1563,43 @@ aggregate_functions:
             values: [ TIE_TO_EVEN, TIE_AWAY_FROM_ZERO, TRUNCATE, CEILING, FLOOR ]
         nullability: DECLARED_OUTPUT
         return: fp64?
+  - name: "first"
+    description: >-
+      First value in a set of rows.
+
+      The `null_handling` option determines whether or not null values will be recognized by the function.
+      If `null_handling` is set to `IGNORE_NULLS`, the function will look for the first non-null value.
+      If set to `ACCEPT_NULLS`, the function will return the first value it sees, even if null.
+
+      If all values are null or there are no values, returns null.
+    impls:
+      - args:
+          - name: x
+            value: any
+        options:
+          null_handling:
+            values: [ IGNORE_NULLS, ACCEPT_NULLS ]
+        nullability: DECLARED_OUTPUT
+        decomposable: MANY
+        intermediate: any?
+        return: any?
+  - name: "last"
+    description: >-
+      Last value in a set of rows.
+
+      The `null_handling` option determines whether or not null values will be recognized by the function.
+      If `null_handling` is set to `IGNORE_NULLS`, the function will look for the last non-null value.
+      If set to `ACCEPT_NULLS`, the function will return the last value it sees, even if null.
+
+      If all values are null or there are no values, returns null.
+    impls:
+      - args:
+          - name: x
+            value: any
+        nullability: DECLARED_OUTPUT
+        decomposable: MANY
+        intermediate: any?
+        return: any?
   - name: "quantile"
     description: >
       Calculates quantiles for a set of values.