Have to insert dummy data for confidences when representing keypoint-moseq results

[kaijfox]

Working on NWB I/O for, keypoint-moseq we're running into an issue because we don't calculate confidences for our denoised pose estimates. The closest thing we have are variances for the Gaussian noise we assume around those re-estimated keypoint locations in our generative model, but that is just a scalar for each keypoint.

We currently have to insert an array of nans in the confidences field - which doesn't feel like a great solution. Do you have a suggestion for a better practice, or would it be possible to make that field optional?

[rly]

On further inspection, it turns out that the "confidence" field in PoseEstimationSeries is already optional as of #11! This has not been released yet (my fault). Please use the main branch of this repo instead of the version on pypi until we can get this new version released.