Use IO-Subgraph + Blockwise throughout codebase

~Blocked by #6715~ ~Blocked by #7281~

After #7281 goes in, most (maybe all) IO operations in the dataframe and array modules can be converted into Blockwise high-level graph (HLG) layers. This change should improve HLG optimizations (like blockwise fusion).

This issue is intended as a “living” checklist of action items related to Blockwise IO…

Dask-Dataframe:

• read_parquet [#7415]
• read_csv (via text_blocks_to_pandas) [#7415]
• read_orc [#7415]
• make_timeseries [#7615]
• daily_stock [#7615]
• read_hdf [#7625]
• from_array
• from_pandas
• from_bcolz
• from_dask_array
• from_delayed
• read_json

Dask-Array:

• from_array
• from_zarr (may be covered by from_array)
• from_delayed
• from_func
• empty_like
• ones_like [#7281]
• zeros_like [#7281]
• full_like [#7281]
• linspace
• arange
• eye
• diag
• diagonal
• triu
• tril
• fromfunction [#7704]
• repeat

[jrbourbeau]

Is it worth creating a doc page explaining how people go about implementing an HLG operation themselves?

Cross referencing https://github.com/dask/dask/issues/7709 and https://github.com/dask/dask/issues/7755 for updating documentation around HLGs

[jakirkham]

Is it worth creating a doc page explaining how people go about implementing an HLG operation themselves? Maybe that helps people feel more comfortable filling out gaps in the codebase. Not to mention writing their own in their code