Allow binding to implement GC trigger

[qinsoon]

This PR allows bindings to create a delegated GC trigger.

• Make GCTriggerPolicy public. Add a new type SpaceStats which simply wraps Space.
• Add Collection::create_gc_trigger. This will be called when the gc_trigger option is set to Delegated.

[wenyuzhao]

LGTM

[wenyuzhao]

Unrelated to this PR: Why is Plan safe to be exposed to the binding, whereas Space is not? (See L126 below in this file)

[wenyuzhao]

Looks like it's still possible to access spaces from a plan reference?

mmtk-core/src/plan/global.rs

Lines 142 to 155 in 7081e86

	pub trait Plan: 'static + HasSpaces + Sync + Downcast {
	/// Get the plan constraints for the plan.
	/// This returns a non-constant value. A constant value can be found in each plan's module if needed.
	fn constraints(&self) -> &'static PlanConstraints;
	/// Create a copy config for this plan. A copying GC plan MUST override this method,
	/// and provide a valid config.
	fn create_copy_config(&'static self) -> CopyConfig<Self::VM> {
	// Use the empty default copy config for non copying GC.
	CopyConfig::default()
	}
	/// Get a immutable reference to the base plan. `BasePlan` is included by all the MMTk GC plans.
	fn base(&self) -> &BasePlan<Self::VM>;

mmtk-core/src/plan/global.rs

Lines 326 to 334 in 7081e86

	pub struct BasePlan<VM: VMBinding> {
	pub(crate) global_state: Arc<GlobalState>,
	pub options: Arc<Options>,
	pub gc_trigger: Arc<GCTrigger<VM>>,
	// Spaces in base plan
	#[cfg(feature = "code_space")]
	#[space]
	pub code_space: ImmortalSpace<VM>,

[qinsoon]

Unrelated to this PR: Why is Plan safe to be exposed to the binding, whereas Space is not? (See L126 below in this file)

Space is not public before this PR, and I don't think we have a good reason to expose it. So I decided to wrap Space into a public type SpaceStats.

For Plan, I don't think we want to expose Plan either (at least some methods in Plan should not be public). Maybe we should check the visibility of Plan's methods, and have another PR to tidy it up.

[qinsoon]

Looks like it's still possible to access spaces from a plan reference?

mmtk-core/src/plan/global.rs

Lines 142 to 155 in 7081e86

	pub trait Plan: 'static + HasSpaces + Sync + Downcast {
	/// Get the plan constraints for the plan.
	/// This returns a non-constant value. A constant value can be found in each plan's module if needed.
	fn constraints(&self) -> &'static PlanConstraints;
	/// Create a copy config for this plan. A copying GC plan MUST override this method,
	/// and provide a valid config.
	fn create_copy_config(&'static self) -> CopyConfig<Self::VM> {
	// Use the empty default copy config for non copying GC.
	CopyConfig::default()
	}
	/// Get a immutable reference to the base plan. `BasePlan` is included by all the MMTk GC plans.
	fn base(&self) -> &BasePlan<Self::VM>;

mmtk-core/src/plan/global.rs

Lines 326 to 334 in 7081e86

	pub struct BasePlan<VM: VMBinding> {
	pub(crate) global_state: Arc<GlobalState>,
	pub options: Arc<Options>,
	pub gc_trigger: Arc<GCTrigger<VM>>,
	// Spaces in base plan
	#[cfg(feature = "code_space")]
	#[space]
	pub code_space: ImmortalSpace<VM>,

BasePlan is not public, though it appears in a public trait. It is the same case for Space in Plan::collection_required before this PR. I am not sure what is Rust's rule on this. There is a lint private_in_public, but it seems Rust does not warn for these cases.