Separate positional from by name cast

[robert3005]

In order to handle full generality of casts we need to distinguish between
casting fields by name and by position. The positional casts are useful for non
sql systems where names are just metadata and the order of values matters.
However, in order to integrate with sql systesm we need by names casts as well.
It's not possible to handle them without any indication from the user which
behaviour they care about

fix #6900

Signed-off-by: Robert Kruszewski github@robertk.io

[robert3005]

Looks like we want to invert defaults here where the cast by default is name based from talking to @gatesn

[codspeed-hq]

Merging this PR will improve performance by 32.81%

⚡ 13 improved benchmarks
✅ 1150 untouched benchmarks
⏩ 1457 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	new_alp_prim_test_between[f64, 32768]	253 µs	193.7 µs	+30.6%
⚡	Simulation	new_bp_prim_test_between[i64, 16384]	145 µs	116.1 µs	+24.89%
⚡	Simulation	new_bp_prim_test_between[i64, 32768]	237 µs	178.4 µs	+32.81%
⚡	Simulation	old_alp_prim_test_between[f32, 32768]	272 µs	241.7 µs	+12.54%
⚡	Simulation	old_alp_prim_test_between[f64, 32768]	351.5 µs	287.2 µs	+22.37%
⚡	Simulation	old_bp_prim_test_between[i32, 32768]	248.9 µs	220.8 µs	+12.76%
⚡	Simulation	old_bp_prim_test_between[i64, 16384]	220.1 µs	189.2 µs	+16.31%
⚡	Simulation	old_alp_prim_test_between[f64, 16384]	238.8 µs	206.9 µs	+15.45%
⚡	Simulation	new_alp_prim_test_between[f32, 32768]	184.8 µs	156.3 µs	+18.2%
⚡	Simulation	new_alp_prim_test_between[f32, 16384]	120.8 µs	106.2 µs	+13.76%
⚡	Simulation	new_alp_prim_test_between[f64, 16384]	157.3 µs	127.8 µs	+23.02%
⚡	Simulation	new_bp_prim_test_between[i32, 32768]	169.8 µs	141.5 µs	+19.97%
⚡	Simulation	old_bp_prim_test_between[i64, 32768]	329.3 µs	266 µs	+23.78%

---

_{Comparing rk/castopts (19ac0da) with develop (24fcc56)}

Footnotes

1. 1457 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[robert3005]

Ok, I will add a cast registry

[robert3005]

@AdamGS I have dug through datafusion issues. The top level one is apache/datafusion#17330 where what Datafusion really wants is to remove cast_column and have cast with cast_column behaviour. So long term datafusion will have one cast. However, Arrow cast is not the datafusion cast given they're migrating away from it.

I will make a pluggable cast

[robert3005]

I will have to redo this with deeper changes. Will close this and open a new pr #7588 is the first step we need for a real solution