Fix issues in sanity GC

[qinsoon]

This PR fixes two issues in sanity GC.

• It fixes Sanity GC is broken #1078 by removing the code to prepare/release mutator/collector, which seems unnecessary.
• It replaces the use of ScanObjects for cached root nodes with ProcessRootNode. Otherwise, the assertion assert!(!self.roots()) will fail in ScanObjectsWork::do_work_common.

[wks]

ScanObjectsWork::do_work_common used to contain code for adding root nodes into the sanity checker. Since 61d20e2, the processing of root nodes (the objects pointed by root edges from outside the object graph, i.e. pointed by stack slots and global variables) is moved to ProcessRootNode, and ScanObjects never processes any "root nodes". Therefore, it is right to replace the use of ScanObjects with ProcessRootNode. (I wonder why 61d20e2 changed the code related to sanity GC but didn't change the actual SanityChecker.)

I also think it is safe to remove the roots() method from the ScanObjectsWork trait, and the roots field from both ScanObjects and PlanScanObjects. The roots() method was only used by the code related to sanity checking. The fact that it is asserted to be false in do_work_common means it is not used now.

[qinsoon]

I also think it is safe to remove the roots() method from the ScanObjectsWork trait, and the roots field from both ScanObjects and PlanScanObjects. The roots() method was only used by the code related to sanity checking. The fact that it is asserted to be false in do_work_common means it is not used now.

I removed roots() in the trait and the fields in its implementations.

[wks]

LGTM

[wks]

There are some example code in the documentation that still uses the roots parameter.

[qinsoon]

binding-refs
OPENJDK_BINDING_REPO=qinsoon/mmtk-openjdk
OPENJDK_BINDING_REF=add-test-for-sanity-gc