Replace ScanOverview.__new__ magic with Pydantic Validator

[pederhan]

Frankly, the use of __new__ on models.ScanOverview is both difficult to reason with and difficult to explain in the docs.

https://github.com/pederhan/harborapi/blob/002ff74abc52293e924d6c91c6aaf44cb9a60dd5/harborapi/models/models.py#L260-L271

This should be changed to make it easier to document and maintain get_artifact(..., with_scan_overview=True), as well as the classes models.Artifact, models.ScanOverview and models.NativeReportSummary.

Why is it like this?

Well, ask the Harbor developers. A ScanOverview seems to contain a dict of mime types as keys, and values that conform to the NativeReport spec. This behavior is not documented. We make an assumption that when a user requests a scan overview, they actually want a NativeReportSummary, not a dict of dicts.

This has the unfortunate side effect of discarding all but one report if there are reports for multiple mime types (seems unlikely?) Such as the case below:

{
    "application/vnd.scanner.adapter.vuln.report.harbor+json; version=1.0": {
        # dict that conforms to NativeReportSummary spec
        ...
    },
    "application/vnd.security.vulnerability.report; version=1.1": {
        # dict that conforms to NativeReportSummary spec
        ...
    }
}

As such, it might be beneficial to rethink the whole approach to parsing ScanOverview, and see if we can come up with something that is easier to maintain and use until this part of the API is properly documented.

Proposed Solution

Unsure. Need to investigate if an artifact can ever have NativeReportSummary dicts for more than one MIME type simultaneously. Can the example above ever occur?