Subtraction overflow in mem balancer implementation

[qinsoon]

If we replace the following subtraction with checked_sub, it may fail.

mmtk-core/src/util/heap/gc_trigger.rs

Line 276 in 0fb520a

self.allocation_pages = (mmtk.plan.get_reserved_pages() - self.gc_end_live_pages) as f64;

I observed this in Julia with non-moving immix:

      From worker 3:    thread '<unnamed>' panicked at 'called `Option::unwrap()` on a `None` value', /tmp/nix-build-julia-mmtk-patched.drv-0/git/checkouts/mmtk-core-89cdc7bf360cce7f/2a6e38e/src/util/heap/gc_trigger.rs:276:100
      From worker 3:    stack backtrace:
      From worker 3:       0: rust_begin_unwind
      From worker 3:       1: core::panicking::panic_fmt
      From worker 3:       2: core::panicking::panic
      From worker 3:       3: <mmtk::util::heap::gc_trigger::MemBalancerTrigger as mmtk::util::heap::gc_trigger::GCTriggerPolicy<VM>>::on_gc_start
      From worker 3:       4: mmtk::scheduler::controller::GCController<VM>::run
      From worker 3:       5: start_spawned_controller_thread
      From worker 3:       6: fn_spawn_controller_thread
      From worker 3:       7: start_thread
      From worker 3:       8: __GI___clone
      From worker 3:    note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
      From worker 3:    fatal runtime error: failed to initiate panic, error 5

I don't know the root cause yet. bdb5dff did not fix this issue.