gdb: send library events during module load

[williballenthin]

Closes #319

When a module is mapped at runtime (for example by LoadLibrary), the gdb stub requests a stop and replies with the library stop reason (T05thread:1;library:;). Clients such as IDA then re-read qXfer:libraries:read, diff the list against what they knew, and fire LIB_LOADED events, so runtime-loaded DLLs appear in the Modules window without a manual refresh. This matches Windows gdbserver semantics: the LoadLibrary handler completes fully (module mapped, PEB updated, return value written) before unicorn halts, and the client resumes on its own unless the user chose to suspend on library load.

Changes

• WindowsEmulator gains module_change_listeners, notified after load_image appends to emu.modules (the single mutation point for the module list). Listener errors are logged, never propagated into emulation.
• GdbServer registers a listener for the session lifetime (added on client connect, removed in close()). A module change sets _library_list_changed and requests a stop through the existing _request_stop machinery.
• _stop_reply appends library:; from that flag alone and clears it, so the field can coalesce with another pending stop reason (a breakpoint reached in the same instruction) or with a dependency chain that maps several DLLs before the halt. One stop covers them all, because the client diffs the full list.
• Reading the flag rather than the stop kind also keeps a repeated ? query from re-reporting an event the client already handled.
• No qSupported change: the library stop reason is not gated by a feature string, and qXfer:libraries:read+ is already advertised.

FreeLibrary does not unmap, so unload events cannot occur. The mechanism is direction-agnostic (clients diff the full list), so LIB_UNLOADED works if module removal is implemented later. doc/gdb.md records both the new stop reason and this limitation.

Testing

• tests/test_gdb_compat.py: a module change requests a library stop and its reply carries library:;; the flag piggybacks on a breakpoint stop reply and is reported once; the listener is removed when the session closes.
• tests/test_gdb.py: end-to-end subprocess test that maps a module mid-run through a code hook, asserts the T05...library:; stop reply, finds the new DLL in the re-fetched library XML, and confirms the next stop does not repeat the event. Suppressing the notification makes this test time out, so it is not vacuous.
• Full suite: 181 passed, 26 skipped. The 3 failures are pre-existing and come from the uninitialized tests/capa-testfiles submodule.

🤖 Generated with Claude Code

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.