listener: stop burning 9 % CPU while nothing is happening

[porterchild]

I was running the listener on a raspberry pi, and it was using enough cpu at idle to interfere with snappiness of other things running on the pi. Hence this optimization.
PR made with @porterchild 's supervision by Qwen 3.8 flash next (running locally, sovereign AI FTW):

A listener holding one idle session uses ~9% CPU and wakes ~100x/s while doing nothing. The main loop paces itself with helpers.SleepRate, but SleepRate.next_sleep_time() built the next wake time from a hard-coded 0.01 instead of self.target_period — the constructor argument was never read, so every SleepRate was a fixed 100 Hz clock. Both callers pass 0.01, which is why it went unnoticed: the knob only fails when you try to turn it.

Why pump_all() has to return something. listener.py already sleeps only when a pump pass sent nothing (if not await pump_all(): await sleeper.sleep_async()), but pump_all() returned None, so that was always true and the loop slept after every pass. At 100 Hz that is harmless; at 0.25 s it means one chunk per 0.25 s, and a send refused by a congested channel only gets retried once per pump pass — 55% of bulk throughput. Returning the flag is what makes throughput independent of the pump rate.

Pump period	pump_all() returns	Bulk, 274 KB
0.01 (HEAD)	no	3.44–3.65 s
0.01	yes	3.44 s
0.25	no	5.49–5.70 s
0.25	yes	3.45–3.65 s

One idle session, interleaved A/B:

	CPU	Wake/s	Session close
HEAD	9.37 %	99	3.56 s
This	1.17 %	14	3.75 s (+0.18)

Chatty session (5 lines/s): 10.7 % → 2.2 %. 0.25 s is the floor — the residue is CallbackSubprocess polling at 10 Hz, and 0.5 s measured identical close latency for no further gain. Worst case it adds is 250 ms before the first byte of output after a quiet gap.

rnsh/initiator.py also uses SleepRate(0.01) and is untouched; with the helper fixed, slowing it down later is a one-line change.

Alternatives considered:
-Event-based instead of polling mechanism. This was significantly more complex and not significantly more responsive or efficient.

Tests: 27 pass. test_rnsh.py fails identically before and after (6 pre-existing failures here), and test_protocol.py is uncollectable before and after (RNS.Channel.ChannelOutletBase).