…igp#448)

We have many hardcoded paths all over the codebase.


  Move all of them and directory creation to a `data_dir` module in global config.

There we can also soon add a lockfile mechanism!

…atabase switchups (sigp#387)

- closes sigp#357


  Introduce the `metadata` table where we track the currently used database schema. This will later allow us to implement "migrations", i.e. updating the database schema without requiring the user to resync.

Furthermore, when we create the database, we store the domain type used in order to avoid working with a database that exists, but was previously used on another network.

The existing `block` table is removed and it's column moved to `metadata` in order to have only one table with data that exists only once.

- sigp#471

This was caused by a bug in `multi_index_map`, see also lun3x/multi_index_map#65.


  Due to time pressure, revert the refactor sigp#463 for now for a smooth `v0.3.0` (and by extension `v1.0.0`) release.

This adds a CLAUDE.md file that provides guidance to Claude Code when working with this repository. The file includes:
- Common commands for building, testing, and linting
- Architecture overview
- Code organization details
- Development tips

🤖 Generated with [Claude Code](https://claude.ai/code)

sigp#437 introduced a bug, where anchor fails to collect signatures for committees with a sync duty.


  This reverts sigp#437.

Also change the incorrect comment, in order to prevent this from being removed again. Sorry!

)

I noticed a fun bug: When in a single batch (for historic sync) or block (for live sync) a validator is removed and readded again, the validator store does not notice the change. This is bad, as the private key share likely has changed. So we continue signing stuff with the old partial share, while correct nodes sign with the new partial share. This causes signatures to be invalid when recombined.


  First, I was tempted to fix this while maintaining the current structure, but I realized that iterating over the whole database every time and comparing it to everything in the validator store is quite nonsensical if we can use the database directly instead.

So this PR gets rid of the `DashMap`s in the validator store and reads needed info directly from the database state.