hw_wallet/qt: reuse message dialog to fix lag between device prompts

[sashazykov]

Signing a transaction with many outputs on a Trezor is noticeably laggy: after I confirm one output on the device, there's a short pause before the next output shows up on the device screen.

What pointed me at Electrum's GUI rather than the device or the library: I have an Android signer (welldone-tech/satoshi-signer) built on the same trezorlib, and it has no such delay — confirming an output brings up the next one instantly. Same library, same device, so the bottleneck is on Electrum's host/UI side.

Symptom

When signing a tx with many outputs, each output confirmation on the device is preceded by a visible pause. On macOS you can see why: the "check your device" popup slides closed and a brand-new one slides open for every output, instead of staying put.

Root cause

QtHandlerBase.message_dialog() tore down the current dialog (clear_dialog() → QDialog.accept()) and built a fresh WindowModalDialog on every device button request. A device emits one button request per output, so signing a tx with N outputs rebuilds the dialog N times. A window-modal QDialog is rendered as an animated "sheet" on macOS, so the popup visibly slides closed and reopens on each request. That GUI churn also competes with the (single) hardware-comms thread for the GIL, adding latency between device prompts.

Fix

Reuse the already-open dialog and just update its label text when one is showing, instead of tearing it down and rebuilding. The window title and on_cancel handler are stable within a signing flow, so the only thing that changes between requests is the message text. If a new request comes in with a different on_cancel, we still rebuild (the existing dialog's cancel wiring would otherwise be wrong).

This also smooths Ledger's repeated progress messages (Signing transaction... (phase2, i/N)), which update the label text on each call.

Testing

Tested on macOS with a Trezor: signing a many-output transaction no longer slides the popup closed/open between outputs, and the lag between prompts is gone — outputs now appear back-to-back.

(I also have a small headless unit test for the dialog-reuse logic if you'd like the suite to start covering this — left out by default, since it would be the first QtWidgets-based test in the tree and pulls in a new QT_QPA_PLATFORM=offscreen setup.)