Remove stale benchmark output files

Implement yield across call boundaries (wippyai#23)

* remove allocator and switch to immutable value boxing

* Add test suite for yield across call boundaries (callR/Call)

Tests cover all code paths that use nested mainLoop calls (callR/Call)
inside the bytecode loop, where yields from Go functions are not
properly propagated:

- OP_TFORLOOP: generic for-loop iterators (7 tests)
- __index/__newindex metamethods (3 tests)
- __add/__sub/__mul arithmetic metamethods (3 tests)
- __concat metamethod (1 test)
- __unm/__len metamethods (2 tests)
- __eq/__lt/__le comparison metamethods (3 tests)
- Mixed boundaries and nested scenarios (3 tests)

All 22 tests fail, confirming the VM limitation. The bug manifests as
either nil pointer dereference (when coroutine state is corrupted) or
double-execution of the iterator (callR returns early on yield but the
calling opcode continues processing with stale register state).

* Implement yield across call boundaries (callR/Call)

System yields (Go functions returning -1) now propagate correctly through
all VM call paths: callR, Call, getField, setField, objectArith, stringConcat,
equals, lessThan, and objectRational.

Each opcode site that calls into user code checks yieldState after the call
returns. On yield, it saves a continuation type (yieldCont) and target register
(yieldContRA), then exits mainLoop. On resume, handleYieldContinuation finishes
the opcode's post-call work (store result, update loop control, adjust Pc).

Yield state merges the old yielded bool and yieldKind into a single yieldState
field (0=none, 1=system, 2=user). coYield returns -2 to distinguish user yields
from system yields without pre-setting state before switchToParentThread.

coResumePropagate handles system yield propagation through coroutine boundaries,
installing a continuation so the next resume re-enters the inner thread.

remove allocator and switch to immutable value boxing (wippyai#22)

test: remove unused tests/app/src/lib fixture

Typecheck performance optimizations round (wippyai#20)

* check: resolve nested table compatibility across manifest refs
* optimize type traversal hot paths for lint perf
* perf: reduce checker inference and traversal overhead
* perf: cache declared symbol ordering in flow solver
* perf: speed resolve symbol typing with cfg scope caches
* perf: cache cfg reverse-postorder between mutations
* perf: reduce allocations in solver point value map
* perf: cut typ traversal and soft-prune overhead
* perf: skip expand-instantiated traversal on leaf types
* test: remove duplicate join returns-or-unknown case
* perf: fast-path soft placeholder pruning on leaf types
* perf: cut typ rewrite/prune builder overhead
* perf: skip visitor dispatch in typ rewrite/prune
* perf: reduce pathkey key construction allocations
* perf: reduce path filter key churn
* perf: add fast pathkey parsers for flow hot loops
* perf: prune mergeFieldAssignments key scans by root prefix
* perf: cache flow point versions and reuse symbol type snapshots
* perf: reuse typ/flow memo maps in hot narrowing paths
* perf: cache resolved path keys during flow narrowing
* perf: speed up constraint solving and condition subsumption
* types: harden table-top narrowing and index semantics

Resolve nested table compatibility across manifest refs (wippyai#19)

Feature/unified fixpoint kernel (syntetic round) (wippyai#18)

Major redundancy reductions, better harness and cleanup.

fix(checker): canonical captured function merge and wrapper regressions

Remove docs folder (wippyai#17)