Performance: Cache config

[cllns]

~28% fewer allocations (138 -> 99) per render
~18% fewer bytes used (8,739 -> 7,160) per render

+36.8% iteration speed (30,891 ips -> 42,266 ips, 32.4μs -> 23.7μs per render)

(Updated these numbers. The early improvements weren't as high but I closed some apps 😇 and these seem to be consistent)

Using benchmarks/view_benchmark.rb on Ruby 4 (on an M1 Pro).

I originally had this with a class-level self.cached_config which isn't necessary so I removed it. This means each instantiation of a view has to compute cached_config for itself, adding some overhead. Since we've built this library based on the assumption that a single View will be re-used (rather than created new each time it's used), I think that's fine. Open to other perspectives though!

Another option is using reader: true for all of the settings. I experimented with this too, but it's not as clean, because the settings are set on the View class rather than the config, so view_class has to be passed around everywhere, which hides the intent of sending just config settings around. It's got basically the same performance as this approach though.

[timriley]

Thanks for looking into this stuff, @cllns! My question is whether we might actually be able to solve this at a Dry Configurable level, rather than create a workaround for it here? We use Dry Configurable in many other places, so a solve there once would benefit us more.

[cllns]

@timriley I was already working on it :) Just had to figure out the right API. See: dry-rb/dry-configurable#167 and updated changes here.

I made a few tiny refactors to move things into private visibility. There's more I want to do, but I'll do that as a follow-on, so this PR doesn't get too complicated.