Fix KnobEventHandler motion drag with log scale

[nabsei]

Fixes #388.

motionEvent() added the linear mouse delta directly to valueTmp and then applied logscale() once. But valueTmp is stored in log-scaled space when usingLog is set (that's what logscale() at the end of each call produces), so adding a linear delta to it and then log-scaling again double-transforms the value — which is why dragging a log-scale knob stayed stuck near the minimum.

The fix converts valueTmp back to linear space via invlogscale() before adding the delta, exactly like the already-correct scrollEvent() a few lines below in the same file (mouse wheel already worked correctly per the original report, and this makes motionEvent() consistent with it).

Verification:

• Wrote a standalone C++ program reproducing the exact logscale/invlogscale formulas and both the old and new motionEvent computations. Simulating 5 drag steps of +1 linear unit each on a log-scale knob (range 1–100, starting at linear value 10): the old code converges to ~2.05 (confirming the reported bug — stuck near the minimum), the new code correctly reaches ~15.0 and matches scrollEvent's math bit-for-bit.
• Built dgl with the Cairo backend (make -C dgl cairo) — compiles cleanly.
• Built the CairoUI example plugin across all formats (CLAP, DSSI, JACK standalone, LV2, VST2, VST3, AU) — all succeed.
• Wasn't able to do a live/visual mouse-drag test in this environment (no display server for interactive GUI testing here), so flagging that the fix is verified mathematically and via successful builds, not an actual on-screen drag.

Small part of this fix (working through the log-scale round-trip math above) was done with the help of Claude Code.