Review Poetry detector tests

Nobody has claimed this yet.

In #184 and #230 we migrated from `Nett` to `Tomlyn` as our TOML parsing library as `Nett` was unsupported and deprecated. We used the `IgnoreMissingProperties` option to allow us to parse `poetry.lock` files which are missing properties defined in our [`PoetryLock`][1] model. Unfortunately, this property was not being respected for objects, like `metadata`.

This issue was fixed in https://github.com/xoofx/Tomlyn/pull/44, and upgrading to `Tomlyn` version `0.15.1` should resolve the issue. However, this _should_ have been caught by [our unit tests][2].

This issue is created to followup on why this was not caught by our tests cases.

See #301 for more context

[1]: https://github.com/microsoft/component-detection/blob/main/src/Microsoft.ComponentDetection.Detectors/poetry/Contracts/PoetryLock.cs
[2]: https://github.com/microsoft/component-detection/blob/b2439a3c67985fb1d13f20e4242eeddd5b65afbc/test/Microsoft.ComponentDetection.Detectors.Tests/PoetryComponentDetectorTests.cs#L26-L55

Start with test/Microsoft.ComponentDetection.Detectors.Tests/PoetryComponentDetectorTests.cs, especially lines 26-55, and read the context in #301. Review how the Poetry lock fixture and assertions cover missing properties such as metadata. Done means identifying why the test suite missed the Tomlyn behavior and updating the tests or documenting the gap so the regression would be caught.

Written by the indexing model from the issue text.