Add tests for core behaviour the suite ran but never checked

[Yaraslaut]

A local Mull mutation run over include/morph/core/ on cc00977 scored 73.0%: 790 mutants, 577 killed, 213 survived. A survivor is a change to the library that morph_tests still passes. This PR goes through all 213. Each one has been applied to the header by hand and given a verdict.

Verdict	Count
Killed by a new test	92
Killed by removing code that had no effect	3
Already caught by the existing suite (Mull misreported it)	42
Cannot change observable behaviour	74
No deterministic way to test	2

Every new test was checked in both directions. It fails with the mutation applied by hand to the header, and passes against the original code.

No library defect was found. Every survivor that changes behaviour was a test gap, not wrong code.

The 42 misreports. Mull reported these as survivors, but each one fails an existing test once the mutation is applied by hand. The likely cause is that Mull mutates one translation unit's copy of an inline header function and the linker keeps an unmutated copy. That cause is inferred, not proven. Either way, survivors reported in header code need checking by hand before anyone acts on them.

The constant 42 means false for a bool. In LLVM IR, Mull's replacement constant 42 truncates to false for a bool. So flag = false → = 42 leaves the program unchanged, and those mutants are counted as unable to change behaviour.

Changes

• Library simplifications. Each removes code whose only effect was already guaranteed elsewhere; behaviour is unchanged.
  • bridge.hpp: a currentId == 0 check that dispatchNow already makes with the same error.
  • wire.hpp: peekCallId's sawDigit flag, which only guarded returning a value that is already 0.
  • remote.hpp: nextOpaqueId's initial value, which was always overwritten.
• TimeoutScheduler documentation. The Handle doc now states that a handle is never 0. DelayAwaiter relies on that.
• Bridge tests:
  • bind and attach supersession across switch and reconnect
  • inline publication when the backend settles on the owner
  • setExecuteDeadline(0) disarming the deadline and starting no timer thread
  • handler-destruction tracking
  • which dispatcher and registry close the registration latch
• Backend tests:
  • a synchronous verb on its own strand runs inline
  • destruction requests stop on a running Task handler
  • cancelled waiters finish as failures
  • executeLatencyMs covers the handler's run time
  • trackedPendingCount
  • shared-instance promotion and release
• RemoteServer and SimulatedRemoteBackend tests:
  • OpaqueIdGenerator is a bijection across both halves
  • drainedWithin(0)
  • an executeTimeout deadline is retired when the reply arrives
  • per-connection release
  • the live-model cap is checked before and around construction
  • the content of the dispatch log lines
  • assignPrimary, private re-bind and cancelPending
• Tests for the other core headers:
  • rollBackShortWrite and repairTornTail edges
  • peekCallId at zero digits and at the uint64 boundary
  • the payload-shape depth limit for every nesting form, plus FNV-1a reference vectors
  • TimeoutScheduler retiring its loop timers on cancel, close and delay
  • the action gate's hand-off after a held leave
  • the thread pool draining its queue on destruction
  • OwnedState::ask posting to the owner
  • observing when no sink was ever installed

The two survivors with no test are remote.hpp's in-flight slot release, which is only reachable if dispatchExecute throws between reserving the slot and posting, and coroutine.hpp:243, a stop racing the timer arm. Neither has a seam that would make a deterministic test.

Re-running the same Mull campaign on this branch, before the rebase so the base is still cc00977, scores 82.78%: 784 mutants, 649 killed, 135 survived. That is up from 73.0%.

• The total is 784, not 790, because the three simplifications removed mutated code.
• Four survivors are new. They sit on lines this branch touched, and none of them can change behaviour.
• Mull still reports 11 mutants as surviving even though a test here was seen failing against the same mutation applied by hand. They have not been re-checked yet. Either Mull's mutation differs from the one applied by hand, or this is the same header misreporting described above.

🤖 Generated with Claude Code

https://claude.ai/code/session_0141t3QdB4FkiHqiCX2M2eF2

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
include/morph/core/remote.hpp	66.66%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!