Poll driver sometimes stuck at 100% CPU utilization in the schedule function

I'm running compio using the polling driver on some old 4.18.X kernel. Sometimes I saw some thread gets stuck in a bad state - it uses 100% CPU all the time, and taking a perf profile shows that majority of the CPU cycle is in the runtime scheduling function.

The workload is roughly the tcp service threads runs a TCP server and spawns work to the source client threads via a flume channel, and it seems to be related to a spike of number of request sent via the cross thread channel.

I'm running compio commit 1341f7253133cd6d9fabd8c6ed0660eedc1120f5, so maybe it's remediated by https://github.com/compio-rs/compio/pull/976?

```
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1233339 root 20 0 8221.4g 27.5g 25.7g R 99.9 1.4 9:35.40 source-tcp-clie
1233340 root 20 0 8221.4g 27.5g 25.7g R 99.9 1.4 9:35.83 source-tcp-clie
1233337 root 20 0 8221.4g 27.5g 25.7g R 99.7 1.4 9:34.02 source-tcp-clie
1233338 root 20 0 8221.4g 27.5g 25.7g R 99.7 1.4 9:33.98 source-tcp-clie
1233489 root 20 0 8221.4g 27.5g 25.7g R 99.3 1.4 9:39.52 cave-tcp-servic
1233519 root 20 0 8221.4g 27.5g 25.7g R 99.3 1.4 9:40.27 cave-tcp-servic
```

Start with the polling driver and runtime scheduling path, then compare the behavior with the changes in pull request 976. Reproduce the high-CPU condition on an older 4.18.x kernel while generating a spike of cross-thread channel requests, and use perf to confirm where the loop remains busy. Done means the scheduler no longer spins at 100% CPU under this workload and the behavior is covered by a regression check.

Written by the indexing model from the issue text.