Fix Dart 3.13 ObjectPool decode: parse early base-object clusters

[fatalSec]

Dart 3.13 dropped the separate VM snapshot. The VM singletons that used to be base objects (empty_object_pool, empty_var_descriptors, the preallocated errors, ...) are now written by CreateEarlyClusters as ordinary clusters at the start of the combined snapshot. On a real 3.13 app this broke the fill walk in two ways.

Three cluster types gained serializers in 3.13 and had no CID kind:

• LocalVarDescriptors: alloc is a count plus one entry count per object; fill is the entry count, one name ref per entry, then per entry four Read<int32_t> (index_kind and three token positions) and one Read<int64_t> scope_id. It is not an inline-bytes class.
• ApiError: fixed-size alloc, one ref.
• UnwindError: fixed-size alloc, one ref plus a bool.

LocalVarDescriptors was read as a simple count, desyncing the stream at cluster 5, so every later cluster, including the global ObjectPool, was left not_parsed and r2flutter -p reported "PP not resolved".

Add the three kinds to the CID model, all version tables and the table generator. Their positions (CompressedStackMaps + 1, LanguageError - 1, UnhandledException + 1) hold in class_id.h from 2.10 through 3.13, and older serializers never emit these clusters, so earlier snapshots are unaffected. Layouts verified against dart-lang/sdk 3.13.0 app_snapshot.cc.

The early empty ObjectPool also meant modern_resolve_pp_slot stopped at the first pool cluster, which is no longer the global pool; skip zero-length pools there.

On the iOS AuthPass 1.9.12 build (Dart 3.13.0, cws=8) all 1140 clusters now fill, -p resolves 44666 entries, PP slots resolve to their targets and -AAA emits valued PP comments. Adds a synthetic r2r test (early-clusters-313) that fails with "PP not resolved" before this change.