Add a macro benchmark suite with longer-running operations, exercising a different set of functionality that is performance-critical.
- One of them could be the speed of an assumeUTXO load-and-dump (this would check block read/write and (de)serialization, but only part of validation).
- Another one could be an actual IBD for 840-880k blocks (small part of IBD, but should produce a representative comparison quickly).
- Another one would be a reindex-chainstate up to e.g. 880k blocks (more realistic, but takes long and doesn't involve reading/writing blocks).
- Lastly, several full IBDs to make sure the previous quick checks are representative (likely only required for big changes). We don't have to run all of these for every PR; it could be tied to labels added to PRs where the author/reviewers think they're relevant (e.g., not for doc typo fixes). This would require integration with CoreCheck.
- Compact block reconstruction