JasperFx 2.80.1, and drop the jasperfx#942 workaround (#423)

[jeremydmiller]

Closes #423.

What's in it

• JasperFx 2.80.1 (from 2.79.2, spanning 2.80.0): all JasperFx packages, plus the AOT smoke's hand-pinned source generator.
• jasperfx#942's workaround is gone. ValueTypeInfo.CreateWrapper / UnWrapper fall back to reflection in a native image since 2.80.0. So ValueTypeDelegates is deleted and StrongTypedIdentification calls ValueTypeInfo directly.
• Compliance: 669 → 678.
  • jasperfx#943 added five GuidOptimisticConcurrencyCompliance facts over a mapped version member. They were skipped on the bump, because they gate on SupportsMappedConcurrencyMember. Fisher has had that route since Guid document optimistic concurrency only works inside one session: a cross-session Store(doc) carrying its version always throws #245, so the fixture now replays MappedVersionMembers and opts in.
  • jasperfx#944's four DocumentSearchCompliance hierarchy facts were green on the bump alone.
• jasperfx#948 (2.80.1, ResilientEventLoader wrapping) needed nothing. Nothing in Fisher matches on EventLoaderException, and the dead-letter and skip suites stay green.

What it does not do: strong-typed aggregate ids in a native image

#423 asked to lift that refusal. It isn't lifted, because #942 fixed only half of it. JasperFxSingleStreamProjectionBase's constructor also calls IEvent.CreateAggregateIdentitySource<TId>() and StreamAction.CreateAggregateIdentitySource<TId>(). Both still CompileFast() for any wrapper id, and that's measured: a natively published readonly record struct PodId(Guid) aggregate throws PlatformNotSupportedException from that constructor on 2.80.1.

Filed as jasperfx#950. The refusal now names #950 instead of #942, and the native smoke asserts it. A statically closed Snapshot<T, TId>() / LiveStreamAggregation<T, TId>() was built and measured, and is parked on a local branch rather than shipped, following #398's precedent of not shipping API that can't do its job natively. It's the one open item of #412.

Verification

• Native smoke (macOS arm64, ILC 10.0.1) passes on 2.80.1. Pinned back to 2.79.2 with the workaround deleted, it fails at ValueTypeInfo.CreateWrapper from StrongTypedIdentification, so the smoke covers the path.
• The mapped-version replay is load-bearing: without it, 4 of the 5 new facts fail.
• Full CI-shaped run (Release, MTP executables with TRX) green on net9.0 and net10.0: 2522 / 36 / 31.
• check_scoreboard.py agrees for both TFMs, and check_no_leaked_databases.py --since is clean.

🤖 Generated with Claude Code