v0.1.20

A follow-up to v0.1.19, from the first run of it on hardware.

An empty page gets three goes at reloading itself, not one

v0.1.19 taught a window that read nothing to retry when the mouse next answered. @dbergPROTAKT ran it twice: once the page filled itself as intended, once it stayed empty while the mouse was demonstrably awake and a manual reload worked straight away (#13).

Both halves of that were real.

One attempt was all it got. A report proving the mouse is awake does not promise its configuration registers answer yet — @f-o-o-g-s measured get_power() timing out on the very push that woke an X3 (#21) — so the single reload could land a moment too early, and the ten-second floor then held the next one off. It now gets three attempts 1.5 s apart, the same shape and spacing as the tray's own adoption read, with the floor applying to the chain rather than to each try.

And only a volunteered battery reading could start one. Whether a mouse pushes those is up to the mouse. What always arrives is a report of some kind, so any report on the listener's node now counts as the mouse being awake, parsed or not — a click or a nudge always lands there. Rate-limited to once a second, because on this family that node is the mouse's real input path and streams while the mouse moves.

---

Found by @dbergPROTAKT on a Pulsar X2 Wireless (Areson), who also confirmed the other two v0.1.19 fixes on hardware: Apply refusing on a page that read nothing, and a mouse plugged in after the app started being picked up.