[FormRecognizer] Test fix and sample reordering

[kinelski]

Set of changes:

• Fixed Ignored sample tests.
• Moved common functionality from FormRecognizerClientLiveTests and FormTrainingClientLiveTests to the new class FormRecognizerLiveTestBase.
• Reordered samples to make training appear before custom recognition.
• "Train Model with Forms" and "Train Model with Forms and Labels" samples numbering updated to be the same.

Fixes #11493

[kinelski]

/azp run net - formrecognizer - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[annelo-msft]

We worked with the other languages decide which samples to expose in what order. I don't think it make sense to change these around and realign with other languages this close to the next release.

[maririos]

Reordered samples to make training appear before custom recognition.

The agreement across languages was to leave custom before training. Users will use custom recognition more often than training. Also, we are expecting for the majority to use the Labeling Tool.

"Train Model with Forms" and "Train Model with Forms and Labels" samples numbering updated to be the same.

Numbering goes according to complexity in the sample. I left them with different numbers because "with forms and labels" implies that you need extra documents with the labeling format and therefore is a way more complicated scenario

[kinelski]

We worked with the other languages decide which samples to expose in what order.

The agreement across languages was to leave custom before training.

I didn't work directly with samples in the last preview, so I wasn't aware it was an agreement across languages. I'm not really fond of the custom => training order, but I agree that deviating from other languages at this point would not be profitable. I'll revert it back.

[kinelski]

Numbering goes according to complexity in the sample. I left them with different numbers because "with forms and labels" implies that you need extra documents with the labeling format and therefore is a way more complicated scenario

I was thinking more about the "code complexity" than the "scenario complexity" itself, since the code of the samples are basically the same, but I don't really mind that much. I think your approach is reasonable as well.

I was a bit concerned about having a Sample4_TrainModel.md file for both Sample4_TrainModelWithForms.cs and Sample5_TrainModelWithFormsAndLabels.cs,

while having Sample5_ManageCustomModels.md for Sample6_ManageCustomModels.cs, but maybe I'm just overconcerning.

I'll revert it back since it's not really this PR's focus. We can always discuss it again if we think it's a reason for concern.

[kinelski]

/azp run net - formrecognizer - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[kinelski]

/azp run net - formrecognizer - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).