Better handle dynamic registry sampler broadcasts

In situations where clients are (dynamically) subscribing *while*
broadcasts are starting to taking place we need to handle the
`set`-modified-during-iteration case. This scenario seems to be more
common during races on concurrent startup of multiple symbols. The
solution here is to use another set to take note of subscribers which
are successfully sent-to and then skipping them on re-try.

This also contains an attempt to exception-handle throttled stream
overruns caused by higher frequency feeds (like binance) pushing more
quotes then can be handled during (UI) client startup.

piker/data/_sampling.py

@@ -253,20 +253,30 @@ class Sampler:
             # f'consumers: {subs}'
         )
         borked: set[tractor.MsgStream] = set()
-        for stream in subs:
+        sent: set[tractor.MsgStream] = set()
+        while True:
             try:
-                await stream.send({
+                for stream in (subs - sent):
-                    'index': time_stamp or last_ts,
+                    try:
-                    'period': period_s,
+                        await stream.send({
-                })
+                            'index': time_stamp or last_ts,
-            except (
+                            'period': period_s,
-                trio.BrokenResourceError,
+                        })
-                trio.ClosedResourceError
+                        sent.add(stream)
-            ):
+
-                log.error(
+                    except (
-                    f'{stream._ctx.chan.uid} dropped connection'
+                        trio.BrokenResourceError,
-                )
+                        trio.ClosedResourceError
-                borked.add(stream)
+                    ):
+                        log.error(
+                            f'{stream._ctx.chan.uid} dropped connection'
+                        )
+                        borked.add(stream)
+                else:
+                    break
+            except RuntimeError:
+                log.warning(f'Client subs {subs} changed while broadcasting')
+                continue
 
         for stream in borked:
             try:
@@ -848,6 +858,16 @@ async def uniform_rate_send(
         # rate timing exactly lul
         try:
             await stream.send({sym: first_quote})
+        except tractor.RemoteActorError as rme:
+            if rme.type is not tractor._exceptions.StreamOverrun:
+                raise
+            ctx = stream._ctx
+            chan = ctx.chan
+            log.warning(
+                'Throttled quote-stream overrun!\n'
+                f'{sym}:{ctx.cid}@{chan.uid}'
+            )
+
         except (
             # NOTE: any of these can be raised by ``tractor``'s IPC
             # transport-layer and we want to be highly resilient

piker/data/feed.py

@@ -1589,6 +1589,9 @@ async def open_feed(
                 (brokermod, bfqsns),
             ) in zip(ctxs, providers.items()):
 
+                # NOTE: do it asap to avoid overruns during multi-feed setup?
+                ctx._backpressure = backpressure
+
                 for fqsn, flume_msg in flumes_msg_dict.items():
                     flume = Flume.from_msg(flume_msg)
                     assert flume.symbol.fqsn == fqsn