oracle: respond with native blocks

[serprex]

depends on ClickHouse/clickhouse-c-rs/pull/1
fixes #110

moves oracle to being used directly by inserters, reducing load on decode pool

[serprex]

Speed:

type	Batch 1 impact	Batch 2 speedup	Batch 64 speedup	Batch 4,096 speedup	Peak branch throughput
int4[] → Array(Int32)	3.8% slower	1.87×	36.0×	80.8×	9.13M rows/s at batch 512
text[] → Array(String)	2.7% slower	1.98×	29.4×	72.9×	5.55M rows/s at batch 4,096
timestamp[] → Array(DateTime64)	9.1% slower	1.66×	29.3×	79.0×	6.44M rows/s at batch 4,096
uuid[] → Array(UUID)	0.4% slower	2.05×	36.3×	69.4×	8.05M rows/s at batch 512
jsonb → JSON	2.1% slower	1.80×	34.8×	109.7×	8.85M rows/s at batch 4,096
Mixed five-column row	19.5% slower	1.74×	17.8×	15.1×	1.43M rows/s at batch 512

Throughput rows/s:

type	Batch 1	Batch 2	Batch 4	Batch 8	Batch 16	Batch 64	Batch 512	Batch 4,096
int4[] → Array(Int32)	70.1k	134k	294k	453k	889k	2.87M	9.13M	6.18M
text[] → Array(String)	71.2k	140k	225k	446k	806k	2.30M	4.52M	5.55M
timestamp[] → Array(DateTime64)	63.3k	111k	210k	441k	796k	2.26M	4.35M	6.44M
uuid[] → Array(UUID)	70.9k	140k	271k	442k	854k	2.54M	8.05M	5.39M
jsonb → JSON	74.5k	133k	269k	440k	887k	2.59M	6.74M	8.85M
Mixed five-column row	48.0k	92.4k	171k	297k	452k	951k	1.43M	874k

[serprex]

needs some pg-clickhouse-c tweaking to support maintaining fine grained error-impacts-one-row rather than error-impacts-whole-batch