Pins the held-scope, precedence and empty-string arms of processor_scope

[johnnyt]

What

Three arms of processor_scope/1 (the scope a send resolves in on the send-processor shape) had no test that went red when they broke. Three tests in test/statifier_router/send_handler_test.exs, describe "the scope a live session names", now pin them:

• With no :processor_scope configured, the scope the calling process holds (SendHandler.put_delivery_scope/1) is the one the send resolves in.
• A configured :processor_scope wins over a held scope, both as a static value and as a fun's answer.
• A fun answering "" is a miss: a send and a delayed send each answer {:error, {:invalid_value, :processor_scope, ""}}, nothing is routed and nothing is queued, even with a held scope in reach.

Tests only: lib/statifier_router/send_handler.ex is unchanged, so no answer changes. No changelog fragment: changelog.d/README.md excludes test changes with no public surface.

Sabotage

Each mutation was applied to processor_scope/1 from a copy of the file, the handler test file was run, and the file was restored byte-equal before the next check:

• the nil-config clause answered {:ok, nil}: "with no scope configured, the scope the calling process holds applies" failed on its assertion;
• the static-scope clause answered the held scope ahead of the configured one: "a configured scope wins over the scope the calling process holds" failed on its assertion;
• the fun clause answered the held scope ahead of the fun's answer: the same test failed on its assertion;
• the fun clause dropped its empty-string guard: "a fun that answers an empty string is a miss, and nothing is routed or queued" failed on its assertion.

Gate

Full mix quality ran green on this exact tree: format, compile with warnings as errors, isolated tests, doc links, dependencies, Credo, docs, tests (399 of 399, 98.4% coverage) and Dialyzer.

Review

In-turn review: I re-read the diff against the bead's acceptance ("a held delivery scope with no :processor_scope configured is used; a configured :processor_scope wins over a held one; a fun answering "" is a miss; each sabotage-proven") and checked each test against processor_scope/1 and override_scope/0 in send_handler.ex. The nil-config clause and the fun's nil answer read the held scope; a binary answer is used as given; any other answer, "" included, is {:invalid_value, :processor_scope, value}. The existing test for an atom answer already calls that tuple "a miss", so the new test uses the same word. Each test cleans up the held scope with on_exit(&SendHandler.delete_delivery_scope/0), as the existing seam test does. The bead's description said the changelog does not name the invalid-value error. That is stale: the 0.5.0 section's lead paragraph names it, so this PR adds nothing there.

Refs: sr-pby