feat(xsnap)!: Update to Moddable SDK 3.8.7

[mhofman]

closes: #7083

Refs: agoric-labs/moddable#9, agoric-labs/xsnap-pub#38 (Will be merged first then this PR updated with the resulting commits)

Description

Update to Moddable SDK 3.8.7, the latest published tag

This includes a change to consider WeakRef as strong during organic gc, mitigating #6784.
This includes an upstream change that grows the property map table as needed, affecting some tests.

Security Considerations

None

Scaling Considerations

None

Documentation Considerations

None

Testing Considerations

Updated xsnap tests to handle the property map allocation change (test disabled), and WeakRef change by switching to a FinalizationRegistry observer instead.

Given the change to WeakRef handling I'm now adapting the acceptance test described in #6929 as follow:

• Replayed from chain transcript using a patched version of XS cherry-picking the WeakRef change, and extracted the vat transcripts used as reference
  • These reference vat transcripts should be insensitive to organic gc and compatible with a wide range of XS versions.
  • Used pismo branch mhofman/6784-hide-organic-gc-pismo which uses agoric-labs/moddable@mhofman/6784-organic-gc-pismo and agoric-labs/xsnap-pub@mhofman/6784-organic-gc-pismo
• Replayed vat transcripts in an integration xsnap-worker updated to 3.8.7
  • replayed with forced reload from snapshot, only keeping the initial never reloaded worker as a baseline
  • verified that all snapshots are identical between reload and initial worker, and that no anachrophobia occurs
  • Used pismo branch mhofman/6784-hide-organic-gc-updated-xs-3-8-7 which uses agoric-labs/moddable@mhofman/update-xs-3-8-7 and agoric-labs/xsnap-pub@mhofman/6784-hide-organic-gc-updated-xs-3-8-7 (which is the xsnap update from fix: keyCount compatibility with updated Moddable SDK agoric-labs/xsnap-pub#38 with a revert of the harden changes that are incompatible with pismo's liveslots)

[warner]

LGTM. I'll be curious to see the memory-usage consequences of the WeakRef changes. Making GC more stable is a good thing, but this will also prolong the lifetime of basically every virtual-object Representative until the next BOYD, which might increase memory pressure, and may encourage us to reduce the reapInterval so they'll happen faster. OTOH, once there's enough headroom, it probably won't matter much.

[mhofman]

All replay tests passed! Merging.