Adding delta compression to Bitpacking compression

[samansmink]

This PR reworks our bitpacking compression implementation, main improvements:

• Add support for FOR-Delta compression (compressing deltas instead of values with FOR)
• Explicit modes for constant and constant delta compression.

The current implementation chooses for each VECTOR_SIZE values the most suitable compression mode.

In the process I also improved the NumericLimits class a bit by making it the Minimum and Maximum methods constexpr.

Compression ratio

TPC-DS SF1 gets roughly 10% smaller from 317992960 to 288108544 bytes.
TPC-H SF1 gets roughly 3% smaller from 264777728 to 256651264 bytes.

Compression ratio validation is in test/sql/storage/compression/bitpacking/bitpacking_compression_ratio.test_coverage. Max compression ratio is like ~800x-900x for constant compression or constant delta compression, delta_for and for can compress up to ~58x-60x.

Performance

Ran some benchmarks on my M1 macbook to validate the implementation.
Firstly added some benchmarks to benchmark/micro/compression to target the different modes:

benchmark	uncompressed	current	this_pr	current vs uncompressed	this_pr vs uncompressed
benchmark/micro/compression/bitpacking/bitpacking_read_constant.benchmark	0.09	0.10	0.11	7%	12%
benchmark/micro/compression/bitpacking/bitpacking_read_constant_delta.benchmark	0.10	0.10	0.11	6%	9%
benchmark/micro/compression/bitpacking/bitpacking_read_dfor.benchmark	0.10	0.10	0.11	2%	14%
benchmark/micro/compression/bitpacking/bitpacking_read_for.benchmark	0.10	0.10	0.10	-0%	3%
benchmark/micro/compression/bitpacking/bitpacking_store_constant.benchmark	2.83	4.21	2.91	49%	3%
benchmark/micro/compression/bitpacking/bitpacking_store_constant_delta.benchmark	3.21	5.46	3.45	70%	7%
benchmark/micro/compression/bitpacking/bitpacking_store_dfor.benchmark	2.85	4.10	3.68	44%	29%
benchmark/micro/compression/bitpacking/bitpacking_store_for.benchmark	3.12	5.44	3.62	75%	16%

then tpch write:

benchmark	current	this_pr	diff
benchmark/micro/compression/store_tpch_sf1.benchmark	12.16	12.00	-1%

Read performance did not change much, bitpacking is already very fast. Adding delta on top doesn't really change much. Write performance has improved quite a bit due to our previous inefficient use of TrySubtractOperator that is now both changed to constexpr and called less often.

Future improvements

Some ideas that are left for future improvements, they're not super important i think:

• emit constant vectors where possible
• support delta compression with nulls
• support delta encoding on unsigned types with values beyond the domain of their corresponding signed type

[Mytherin]

Thanks for the PR! Great work and great results! Some comments below:

[Mytherin]

Thanks!