PYTHON-1752 bulk_write should be able to accept a generator

[sleepyStick]

Notes:

• I only did this for collection.bulk_write, if we like how it was done, i'd be happy to replicate the change for client.bulk_write, but didn't wanna make the change until I knew we liked how it was done here
• if a user passes in a generator and doesn't care about ordering, i currently still consume the generator to sort the group the requests by type...the user can pass in a generator and insist on original ordering which wouldn't consume the generator. I was thinking of maybe changing this so that we only consume part of the generator (like maybe up until max_batch_size instead of the whole generator but this seemed to require creating an _AsyncBulk class for each batch which felt...wrong? I was probably going about this incorrectly? Not sure.
• the test that i picked to duplicate to check generators was a bit arbitrary, but it felt like a good one to pick. If we want more, I'm happy to do so and open to suggestions for which tests specifically.

[ShaneHarvey]

I think the goal of this ticket is to avoid inflating the whole generator upfront and only iterate requests as they are needed at the encoding step. For example:

coll.bulk_write((InsertOne({'x': 'large'*1024*1024}) for _ in range(1_000_000))

If we inflate all at once like we do here, then that code will need to allocate all 1 million documents at once.