Requests with very large payload (>1Gb) need to be gracefully handled

[Ectelion]

Describe the bug
Currently, there seems to be no validation that gracefully handles the case with very large payload (>1Gb are possible). Regardless of payload size, the Broker attempts to load it in-memory, which in some cases makes it prone to out-of-memory issues.

Expected behavior
There should be a cap on payload. The Broker shouldn't attempt to process requests with payload exceeding the cap, but raise a client error (i.e. a response code like 4xx). It probably makes sense to make this cap configurable.

To Reproduce
Both of the following tests resulted in a repro, although this must also be reproducible under a lower RPS.
Test 1: Generate the load of ~100 RPS with 3.5Gb request payload.
Test 2: Generate the load of ~1000 RPS with 256Mb request payload.

Exit critera
A unit test validating this behavior.

[Ectelion]

I think the max cap should be lower than both of (1) default Pub/Sub throughput quota and (2) OustandingBytesLimit on ReceiverSettings (otherwise, the message may get stuck on PubSub).

[Ectelion]

Actually, same issue was just reported in #1753, so we may consider this as a duplicate.

[Ectelion]

There is actually a bit wider issue: down the page with PubSub quotas, there appears to be an additional table with non-extendable quotas. It specifies "10Mb" as the max message size. Just in case, I've double checked and requests with payload above that are indeed not passing at the publishing stage.