[Optimizer] Use OptimizationResult rather than OptimizationState in more places

[serenity4]

This is a data-flow refactor that removes the need to carry OptimizationState around as much, especially during finish! and in result.src for inlining. Instead, more fields have been defined for OptimizationResult to hold the minimal amount of information needed to perform these possibly required transforms:

• compute_inlining_cost!, relying on .inlining_cost::InlineCostType and .rt::Any.
• ir_to_codeinf!, relying on src::CodeInfo and code_info::CodeInfo. Strictly speaking, only one field is required as ir_to_codeinf!(src, ir) mutates src in place, but I'd rather avoid ambiguity.

With this in mind, transform_result_for_local_cache and transform_result_for_cache should now return an OptimizationResult, and result.src should no longer be of type OptimizationState once finish! returns.

The next step (in a future PR unless it is deemed preferable to do it in this one) would be to try to replace the volatile inference mechanism introduced in #51934 with OptimizationResult by using its ability to hold IRCode without requiring a round-trip to CodeInfo upon retrieval for inlining passes.

[serenity4]

@Keno are you good with this direction in terms of data-flow? That should be in line with what we already discussed, but I had to grow OptimizationResult a bit further to avoid having weird couplings with OptimizationState and InferenceResult.