share/availability/light: Unmarshalling of failed sample results in false-positive error

[renaynay]

2024-11-15T15:58:13.262Z	ERROR	das	das/worker.go:134	failed to sample header	{"type": "retry", "height": 3089050, "hash": "F2D6B3C115227531BB89B525073451475ABA4D997445043622A7A336196218F8", "square width": 2, "data root": "8F86D0659F5D39A08F3064E0352F64C5EA1558AA8C607A428A8722108268F090", "err": "invalid sampling result: expected 16 samples, got 4", "finished (s)": 0.00136031}
2024-11-15T15:58:13.262Z	ERROR	das	das/worker.go:134	failed to sample header	{"type": "retry", "height": 3089855, "hash": "C6805A02F90C480BA5422D419BDFCA044427F6CE6A2BAF34EB0B6DA18190D035", "square width": 2, "data root": "DAD2887B6E3BC2FB4E27ED6ADBD350744CB1121AA3C9E2F60D87BE57522517E7", "err": "invalid sampling result: expected 16 samples, got 4", "finished (s)": 0.00147832}

with corresponding non-empty block:
https://mocha.celenium.io/block/3089855?tab=transactions

This is because of https://github.com/celestiaorg/celestia-node/blob/main/share/availability/light/availability.go#L106

While we allow for smaller sampling set when generating the randomised coordinates, we still throw error if it doesn't align with the default amount (16). So if some retries occur on a block that looks like this, this error will happen.

[renaynay]

On second thought, we shouldn't sample all 4 quadrants of the square - one is sufficient and should be considered as an optimisation.