dashboard: fetch fragment graph for each streaming job on demand

[BugenZhao]

Under current implementation, we fetch the fragment graph for all streaming jobs at once as soon as user navigates to the "Streaming Graph" page. This will lead to problems if there are too many jobs in the cluster:

• dashboard: fragment graph loads slowly when there's limited bandwidth and too many actors #15127, which is workarounded by feat(dashboard): enable compression for API & minor refactors #15558
• the web page may respond slowly and even crash due to high memory usage (which I encountered recently when debugging a customer's cluster)

I would suggest fetching the graph for each streaming job on demand to thoroughly resolve this issue. Note that this requires some more refactoring on the layout and back-pressure visualization as well.

[BugenZhao]

Is this addressed by #18272? @kwannoel