input stops working when rendering too often

Looks like a bug or a race, but I can't pinpoint the exact reason. Consider the following snippet:

```
package main

import (
"flag"
"fmt"
"log"

ui "github.com/gizak/termui/v3"
"github.com/gizak/termui/v3/widgets"
)

type state struct {
pressed uint
draw uint
}

func prompt(st *state) {
st.draw++

ui.Clear()
p := widgets.NewParagraph()
p.Text = fmt.Sprintf("[%v] pressed %v times", st.draw, st.pressed)
p.SetRect(0, 0, 25, 5)

ui.Render(p)
}

func main() {
flag.Parse()

if err := ui.Init(); err != nil {
log.Fatalf("failed to initialize termui: %v", err)
}
defer ui.Close()

st := &state{}

prompt(st)
events := ui.PollEvents()
for {
e := <-events
st.pressed++
if e.Type == ui.KeyboardEvent && e.ID == "" {
break
}
prompt(st)
}
}
```

I'm running this on Linux, At a point where I reach ~80 key presses, the input stops working. I've written a small snippet that uses only `termbox-go` to read input and print the state, and the input does not block. Using `ui.PollEvents()` and rendering test through `termbox-go` works well too. However, combining `ui.Render()` and `ui.PollEvents()`, blocks after registering some number of input events.

Start with the provided Go reproducer and inspect the interaction between ui.Render() and ui.PollEvents(). Compare it with the termbox-go-only input test and the version that polls events without rendering; done means repeated rendering continues accepting keyboard input beyond roughly 80 presses.

Written by the indexing model from the issue text.