fix Linux segfault from stale JAWT drawable on OpenGL backend

[jamesarich]

Fixes #564

The segfault inside render() was caused by incorrect JAWT usage: Display* and Drawable were captured once in JawtContext's constructor and cached forever. The JAWT spec requires re-fetching surface info after every Lock() call — the X11 Drawable (window ID) changes when the AWT peer is recreated on first show, on resize, or when JAWT_LOCK_SURFACE_CHANGED is returned. Passing a stale Drawable to glXMakeCurrent/glXSwapBuffers is what causes the crash.

Raised as one of the items in @sargunv's comment at sargunv/maplibre-native#2 (comment).

Changes

• jawt_context.hpp — lock() now calls GetDrawingSurfaceInfo() after each Lock() and refreshes platform surface info (display, drawable, etc.); unlock() calls FreeDrawingSurfaceInfo() before Unlock(), following the required JAWT lifecycle. Constructor simplified to a single lock/unlock to initialise the Display* (which is the JVM's persistent X11 connection and never changes).
• canvas_opengl_backend.cpp — destructor: replace check() (throws from a destructor → UB during stack unwinding) with an early-return guard; use glXMakeCurrent(display, None, nullptr) to release the context without needing a valid window; activate() checks the glXMakeCurrent return value and throws a descriptive error on failure rather than silently proceeding with no current context; deactivate() also uses None as the drawable when releasing, for consistency.
• canvas_frontend.cpp — remove the TODO comment that flagged the segfault.

Checklist

To your knowledge, are you making any breaking changes? No.

Have you tested the changes? On which platforms?

• macOS: builds and runs cleanly; cannot reproduce the Linux segfault on macOS but the fix is deterministic from the JAWT spec.

[sargunv]

I've also prompted AI to help investigate this, and it loves to flag this, but the segfault persists on Linux. This bug requires some deeper exploration (and actual testing on a Linux machine)