add node type indexing

[sunya-ch]

This PR is related to the issue #216 and TODO list in previous PR #222.

The main change is to add node_type indexing in src/train/profiler/node_type_index.py.
I introduce NodeTypeSpec, NodeTypeIndexCollection to group inputting machine data by spec on each pipeline training.
As shown below, at collection, we will autogenerate NodeTypeSpec (processor, #cores, #chips, memory and so on) and keep it in data path. At training, we will read that value by the machine id and then index it in the NodeTypeIndexCollection.
If the same spec has been indexed, it will use the same index number. However, we expect a step to append data from the same group before training. For AWS instance, we expect single profile per one index. The machine index will be kept under pipeline folder in Json format (node_type_index.json). We can read this file and generate machine index on export.

In addition to above enhancement, this PR also includes multiple bug fixes on CI workflow including adding complete-train pipeline run on tekton test.

Signed-off-by: Sunyanan Choochotkaew [email protected]

[sunya-ch]

I will update exporter for separating each node type. Here are examples of exported value for the pipeline trained on SPECPower data.

https://github.com/sunya-ch/kepler-model-db/blob/specpower/models/v0.7/README.md

Pipeline README page

Model error report page (per node_type)

[rootfs]

also cc @KaiyiLiu1234