7e900b7e

A rule that sees a whole package

Replaces #7, which GitHub will not let me reopen: I merged it by mistake and
then took it back off master.

Stacked on #13, which is where the typed paths this uses come from. Review that first; this diff is against it.

Adds `HsGenValidInGenPackage` from `haskell-style.md`, the first rule that
cannot answer from one module, and with it the fact store that lets a rule see
further.

46 files, 1,509 lines. 240 tests.

## Why a database

What a rule above the module level does with facts is join them. One table per
fact, owned by the rule that writes it, so a rule brings its own schema the way
it brings its own check. The envelope never learns what is in that table, which
is why adding a rule adds no case to it: a package rule is a migration, what it
writes out of one module, and the query it answers with.

A fact is a row rather than a keyed entry. Two instances in one module are two
facts, and there is nothing to key them on that is not invented.

Where the store lives is a two-constructor type rather than a string, because
`:memory:` is not a file name: it is SQLite's word for having no file at all.

## The level

A rule's constructor is also its level, so nothing can disagree with it. Each
level judges exactly the suppressions naming its own rules, which is what keeps
unused-suppression detection sound without a global pass over the repository.

## Still not here

One process, so the store has nowhere to go and nothing crosses a process
boundary. Nothing reads what the compiler wrote down: the package check's
signature does not take the compiler's answers, because this rule has no use
for them.

weeder removed seven store queries on the way in that only a project rule would
want.

## What arrives with the store

The store is what needs a fact written down, so the serialisation arrives with it
rather than sitting unused one PR earlier:

- `persistent`, `persistent-sqlite`, `esqueleto`, `path-pieces` and
  `http-api-data`, and the `PersistField`/`PersistFieldSql` instances on the name
  and place types. #10 opens no database and carries none of this.
- `parseSpan` and `parseModuleRef`, which exist so a column can be read back.
- The `HasCodec` instances on the fact types a column holds whole, which are the
  ones nothing queries into.
- `componentModelDeclaredModules`, which #10 assigned and never read.

## Review loop

```
nix flake check
nix develop --command cabal test hopinion-test --test-options="--ai-executor"
```