[mmarusiak]

Summary

While using the macOS build for a cryptanalysis course (lab on brute force and RSA lattice attacks) I ran the
Factorization, Prime Number Generation, Factoring Knowing a Fraction of p, Attack on Stereotyped Messages,
Attack on Small Secret Exponents (Bloemer/May) and Brute-Force Analysis dialogs and hit several problems.
This PR fixes them; each fix is a separate commit.

Fixes

1. Deadlock in the factorisation progress dialog (kernel.cpp, wnd.cpp, internal.h).
   CDlgProgressFactorisation::OnInitDialog spins in Sleep(2) on the main thread until all worker threads have
   registered, but a worker that already holds the dialog lock waits in RunOnMainThread (via SetTimer ->
   IsWindow) for the main thread. Sleep() now processes pending events when called on the main thread.
   "Complete factorization into primes" froze the whole app before.
2. Stack overflow when pasting a long modulus (controls.cpp, SetControlText).
   Both branches called SetValue, so setting an edit to its current text still fired a change event. Handlers that
   update sibling edits (stereotyped-message dialog) recursed until the stack overflowed (SIGSEGV in
   OnChangeEditplainfull). Only notify when the text really changes.
3. Group boxes swallowed mouse clicks (controls.cpp).
   Win32 group boxes are transparent to the mouse, but the rc files often declare them after the controls they
   enclose. A wxStaticBox created later sat on top, making edits and radio buttons (e.g. the ciphertext field of the
   stereotyped-message dialog) unclickable. Group boxes are now kept at the bottom of the z-order.
4. Document tabs could not be selected or closed (docview.cpp).
   The wxAuiTabCtrl id falls into the command id range, so UpdateCommandUI treated it as a menu command without a
   handler and disabled it (MFC auto menu enable). The tab control and notebook are now skipped.
5. Bloemer/May dialog erased N, e and d while typing (RSABloemerMayDlg.cpp).
   checkStartButton updates the bit length and delta edits, whose change handlers clear N, e and d (meant for user
   edits). With change notifications on programmatic updates, typing a digit in N wiped it. The handlers now ignore
   updates made by the dialog itself.

Testing

Built on macOS (arm64, Homebrew wxwidgets@3.2 openssl@3 gmp ninja cmake) and exercised manually:
factorisation of an 18-digit number, prime generation up to 10^700, all three lattice-attack dialogs (results
cross-checked against an independent implementation), AES/IDEA brute-force analysis with unknown key bits, and
opening, switching and closing several document tabs. No automated tests were added.

Known issue (not fixed here)

Generating two random primes of about 1200 digits in "Prime Number Generation" crashes with a segfault in
GeneratePrimes::MillerRabinTest (null Big in compare, worker thread). Ranges up to 10^700 work. I did not find
the cause and do not know whether it is port-specific.