fix: preserve forwarded mouse DPI metadata during injection

[mizuikki]

Summary

This fixes a pointer feel regression when injecting mouse devices.

Before this change, the forwarded virtual mouse created by input-remapper
could lose hwdb-derived properties such as MOUSE_DPI. On devices like the
Logitech G Pro, that caused the injected pointer to feel different from the
original device even though the event capabilities were copied.

What changed

• keep the forwarded device name identical to the source device so system
  mouse hwdb rules can still match
• assign a dedicated input-remapper/... phys path to forwarded devices so
  they remain distinguishable without reusing the original hardware phys
• add a udev rule that restores ID_BUS for forwarded devices before
  70-mouse.rules runs, allowing MOUSE_DPI to be applied
• prevent autoload and device discovery from treating input-remapper-created
  devices as real hardware
• update tests and local uinput mocks to cover forwarded phys handling and
  input-remapper device filtering

Why

On my system, enabling injection for a Logitech G Pro changed the effective
pointer curve/DPI feel. Inspecting the injected device showed that the
forwarded virtual mouse was missing the original device's MOUSE_DPI
metadata.

After this patch, the forwarded device keeps the original device name, gets
ID_BUS=usb, and successfully receives the same MOUSE_DPI value as the
physical mouse.

Verification

Manual verification on Linux with a Logitech G Pro:

• before the patch, the forwarded device had no MOUSE_DPI

• after the patch, both the physical device and the forwarded device expose:

  MOUSE_DPI=400@1000 *800@1000 1600@1000 3200@1000 6400@1000

• pointer feel is now indistinguishable during injection in real use

Notes

I also observed unrelated instability in the local test environment on
Python 3.14, but the DPI/metadata fix itself was validated on the real
device.

[sezanzeb]

Thank you!

There are a few old open issues about the mouse pointer speed, are they fixed with this?

• #358
• #454
• #1048
• #1153

[mizuikki]

Thanks! I addressed the two inline comments in 89709c3:

• UInputMock now exposes phys as an explicit argument.
• The forwarded-device discovery test no longer uses include_inputremapper=True, since that option is about filtering already-discovered input-remapper groups. Devices with an input-remapper phys marker should be skipped during discovery.

All checks are passing on the latest commit.

About the linked pointer-speed issues: I would not claim this PR fixes all of them yet. This specifically fixes the case where the forwarded virtual mouse loses hwdb-derived metadata such as MOUSE_DPI, because the virtual device no longer matches the original mouse hwdb rules. That matches the issue I reproduced locally with my Logitech G Pro.

For reports caused by missing MOUSE_DPI on the forwarded device, this should help. Some of the linked reports also involve KDE/Wayland/X11 pointer speed, acceleration, or per-device settings not being applied to the forwarded virtual device, so those still need confirmation from affected users before closing the issues.

[sezanzeb]

I think include_inputremapper can be removed from both filter and find

[mizuikki]

Done, thanks. I removed include_inputremapper from both filter and find in 988e24c.

Since input-remapper-created devices are now skipped during discovery, the extra flag is no longer needed there. I also updated the reader client and group tests to use the simplified API.

5251c11 only fixes the resulting black formatting issue in reader_client.py. All checks are passing on the latest commit.

[sezanzeb]

I just tried your changes locally, and mouse speed correctly stays the same (Kubuntu 25.10). I am happy about the code comments and that you also worked on the tests.

Just one more thing, probably: Now that filter doesn't filter anything anymore, lets rename it to get_groups. Then I think I'll merge this.

[mizuikki]

Done, thanks. I renamed filter() to get_groups() in e82807b and updated the GUI callers and tests.

All checks are passing on the latest commit.