[Ray] Optimize ray executor submit subtask

[fyrestone]

What do these changes do?

There is a set intersection logic in Ray executor to determine whether the subtask result chunks are the stage results. It could be very slow when the stage has too many chunks and subtasks, e.g. a stage with 100000 result chunks and 50000 subtasks. This PR reuses the info generated in constructing a subtask to avoid set intersecion.

Also, Fix Ray executor track bug.

Test code

join_key_range = 100
data_size = 5000000
chunk_size = 100

mars.new_session(backend="ray")

df1 = md.DataFrame(
    mt.random.randint(
        1, join_key_range + 1, size=(data_size, 10), chunk_size=(chunk_size, 5)
    ),
    columns=list("ABCDEFGHIJ"),
)

start_time = time.time()
result = df1.execute()
end_time = time.time()
print(f"Finished in {end_time - start_time} seconds and execution result: {result}")

This case submit 50000 subtasks in a stage.

• total time, before: 211.34s, after: 87.67s
• submit subtask time, before: 153s, after: 35s

Related issue number

Fixes #xxxx

Check code requirements

• tests added / passed (if needed)
• Ensure all linting tests pass, see here for how to run them

[qinxuye]

LGTM

[chaokunyang]

LGTM