Fix race in hackney_pool in hackney 4.2.1

[kpy3]

In some cases on flaky network connection can die after livness check with noproc, which crashes pool.

=CRASH REPORT==== 5-Jun-2026::19:26:39.618340 ===
  crasher:
    initial call: hackney_pool:init/1
    pid: <0.1206.0>
    registered_name: []
    exception exit: {noproc,{gen_statem,call,[<0.1427.0>,is_ready,infinity]}}
      in function  gen:do_call/4 (gen.erl:249)
      in call from gen_statem:call/3 (gen_statem.erl:3254)
      in call from hackney_pool:find_available/2 (./project/_build/default/lib/hackney/src/hackney_pool.erl:782)
      in call from hackney_pool:handle_call/3 (./project/_build/default/lib/hackney/src/hackney_pool.erl:487)
      in call from gen_server:try_handle_call/4 (gen_server.erl:2470)
      in call from gen_server:handle_msg/3 (gen_server.erl:2499)
    ancestors: [hackney_sup,<0.1012.0>]
    message_queue_len: 48

It looks like a race, where connection dies between calls to it. This fix tries to eliminate crash by forcing creation of new connection in case when no connection with given Pid already exists.

[benoitc]

Reviewed and verified. I reproduced the crash (pool dies on the noproc from is_ready/1 when a pooled connection dies right after the is_process_alive/1 check) and confirmed this fix resolves it. The try/catch matches the existing defensive calls in this module (stop_conn/1 and the is_upgraded_ssl/is_no_reuse checks in do_checkin). Follow-up with a regression test and a small whitespace/comment tidy coming in a separate PR.

[benoitc]

Thanks for the patch ! Hackney 4.2.2 is out with it :)