RecursionError in execute_concurrent when futures complete synchronously

[dkropachev]

Description

execute_concurrent can hit Python's recursion limit and raise RecursionError when futures complete synchronously (e.g. due to very short request timeouts).

Root Cause

When a ResponseFuture already has _final_exception set by the time add_callbacks() is called (e.g. the timeout expired synchronously inside send_request()), the errback fires inline. This creates an unbounded recursion:

_execute → execute_async → add_callbacks → errback fires synchronously →
  _on_error → _put_result → _execute_next → _execute → ...

The existing _exec_depth / max_error_recursion guard only protects the except Exception branch (when execute_async itself raises). It does not protect this synchronous-callback path, so the recursion depth equals the number of remaining statements and blows the stack.

A secondary issue: ConcurrentExecutorListResults._results() uses sorted() on (idx, ExecutionResult) tuples. If comparison reaches the exception objects inside ExecutionResult, it raises TypeError because exceptions don't support <.

How to Reproduce

Use execute_concurrent with a very short request_timeout (e.g. 0.005s) and many statements:

from cassandra.cluster import Cluster, ExecutionProfile
from cassandra.concurrent import execute_concurrent

profile = ExecutionProfile(request_timeout=0.005)
cluster = Cluster(execution_profiles={'short': profile})
session = cluster.connect()

statements = [("SELECT * FROM system.local", ())] * 2000
# This may raise RecursionError or TypeError
results = execute_concurrent(session, statements, raise_on_first_error=False,
                             execution_profile='short')

Observed in CI

FAILED tests/integration/standard/test_query.py::LightweightTransactionTests::test_no_connection_refused_on_timeout
TypeError: '<' not supported between instances of 'RecursionError' and 'NoHostAvailable'