Fix failure when computing Snowflake distinct array_agg

This changes how Snowflake array_agg is implemented.
Before the change it was implemented via 'simplify', but this has the
limitation that aggregated values need to be the same as sorting values
when distinct is on (otherwise results would not be well-defined). The
new implementation delegates at the accumulator level and casts to
VARIANT at the aggregation finalize stage.

For this to work, some changes are backported in DataFusion fork.

GitOrigin-RevId: 5bf1370d844aba5df3fa59d7ddc2b5a4efc84bab

Author: findepi

Cargo.toml

@@ -44,7 +44,7 @@ version = "2.0.0-beta.28"
 # numbers as crates.io / the upstream repository.
 # See https://doc.rust-lang.org/cargo/reference/overriding-dependencies.html#the-patch-section
 [patch.crates-io]
-datafusion = { git = "https://github.com/sdf-labs/arrow-datafusion", rev = "fba71744de5a40aed493afb3ef9b810a89cbe6fc" } # @ sdf/45 branch
+datafusion = { git = "https://github.com/sdf-labs/arrow-datafusion", rev = "7fb794e2776d0601328c359aaa6a47c8d94deffc" } # @ sdf/45
 # OR, when working locally
 #datafusion = { path = "../arrow-datafusion/datafusion/core" }