Simplify table client APIs

[begelundmuller]

Summary

The client APIs for dealing with tables has gotten a little too nested and intricate. This issue investigates if we can settle on a flatter and more straightforward interface. Ideally, we can unify the lower-level and "easy" functions in one interface. It seems it's okay to be slightly less verbose than we are today (the verbosity might actually be adding complexity).

Risks and challenges

• Balance between verbosity and simplicity
• Is it a sufficient leap in simplicity?
• Make sure it's portable to JS and the upcoming Go client

Involved components

• Python client
• Data server

Proposed changes

WIP alert! Starting with the UX, I imagine a refactored Python client interface like this:

table = await client.table("user/project/table", schema=...)
table = await table.version(...) # optional

# writing
await table.write(records, deletes=...)

# querying
cursor = await table.log(replay=True, changes=True, subscription=...) # consumers embedded
cursor = await table.index("time >= '2021-02-01'")
cursor = await client.olap(f"SELECT count(*) FROM {table}")

# reading
df = await cursor.df()
records = await cursor.all()
async for record in cursor:
    print(record)
async for batch in cursor.batches(size=...):
    print(record)
await cursor.consume(callback)
# + consumer APIs, including batching and time delay

# get/set features
await table.get({ ... }) # maps to table.index
await table.set({ ... }, { ... }) # oltp writes (+ ability to pass timestamp to fence writes)
await table.delete({ ... }) # oltp delete

# checkpointers?