perf: Improve median with no grouping by 2X

[2010YOUY01]

Which issue does this PR close?

This is a small (one 20-lines change) follow-on to #13681

Rationale for this change

@Rachelint (in #13681) makes median() aggregate function significantly faster by implementing GroupsAccumulator
However, regular Accumulator is still used in no-grouping aggregates and window expressions, this PR took some code from #13681 to also improve median()'s regular Accumulator implementation

Benchmarks:

DataFusion CLI v44.0.0
> CREATE EXTERNAL TABLE IF NOT EXISTS lineitem (
        l_orderkey BIGINT,
        l_partkey BIGINT,
        l_suppkey BIGINT,
        l_linenumber INTEGER,
        l_quantity DECIMAL(15, 2),
        l_extendedprice DECIMAL(15, 2),
        l_discount DECIMAL(15, 2),
        l_tax DECIMAL(15, 2),
        l_returnflag VARCHAR,
        l_linestatus VARCHAR,
        l_shipdate DATE,
        l_commitdate DATE,
        l_receiptdate DATE,
        l_shipinstruct VARCHAR,
        l_shipmode VARCHAR,
        l_comment VARCHAR,
) STORED AS parquet
LOCATION '/Users/yongting/Code/datafusion/benchmarks/data/tpch_sf10/lineitem';
0 row(s) fetched.
Elapsed 0.011 seconds.

> select median(l_orderkey) from lineitem;
+-----------------------------+
| median(lineitem.l_orderkey) |
+-----------------------------+
| 29993285                    |
+-----------------------------+
1 row(s) fetched.
Elapsed 0.529 seconds.

PR: 0.5s
main: 1s

What changes are included in this PR?

Changed state() implementation in median's Accumulator

Are these changes tested?

existing tests

Are there any user-facing changes?

No

[Rachelint]

Seems we can take all_values directly, and reduce the allocation?

[2010YOUY01]

updated, thanks!

[Rachelint]

Thanks @2010YOUY01 , it is really good.

[alamb]

🚀