Linter service clears lint results too greedily

### Bug Description

When implementing a full lint lifecycle, both pre and post linters are run on an artifact. This requires `LinterService.run()` to be called twice, since each stage must be invoked individually (`LinterService.run(Stage.PRE, ctx); LinterService.run(Stage.POST, ctx)`).

This means:
- Lint results are broken up in a manner that appears arbitrary to users; pre/post really are implementation details.
- `LinterService.summary()` forgets the results of the first run, which could result in false positives for `0` return codes.

### To Reproduce

Implement two custom linters, one pre- and one post-, and then try to run both of them as you might during a pack command's final steps.

### part yaml

```shell
N/A
```

### Relevant log output

```shell
N/A
```

Start with LinterService.run() and summary(), then trace how Stage.PRE and Stage.POST are invoked during the pack command. Reproduce the issue with one custom linter for each stage; done means results from both runs remain available to summary() and do not produce false positives for zero return codes.

Written by the indexing model from the issue text.