test(dig-wallet): make the sweep guard's worked examples doctests — a backwards example survived four gate rounds

[MichaelTaylor3d]

Task

The worked examples in crates/dig-wallet/src/sage/sources.rs's sweep documentation are prose, not
executed code. Make them doctests, so a wrong example fails CI instead of misleading a reader.

Why this is worth doing

A doc example that was exactly backwards survived FOUR gate rounds on that file. :501 claimed
#[cfg(test)] fn f() { ChiaQuery::new(c); } fails loudly; it measures sites=[2] — the construction
is dropped. Correct behaviour (it is test code), and the doc asserted the opposite.

Four rounds of adversarial review read past it, because reviewers read a doc example as documentation
rather than as a claim to test. Everything else in the file was measured; only the examples were not,
and that is precisely where the false statement lived.

The same file also carried, across earlier rounds, "It fails when a second ChiaQuery::new appears in
production code anywhere in this crate" and "every imprecision fails LOUDLY" — both false, both
retracted. The pattern is that the guard's prose over-claims and nothing mechanically contradicts it.

Scope

• Convert the worked examples in the sweep / production_lines documentation into doctests that
  actually run the classifier and assert the measured sites / ended_inside_a_test_item values.
• Keep the inline measured outputs that are there now — they are correct as of c5aad53; the doctest just
  makes them self-enforcing.
• A doctest that cannot fail is worse than prose, so each one must be shown to fail when its expected
  value is altered.

Out of scope

Do not add another classifier heuristic. Five gate rounds established that the durable artifact is the
fail-closed flag plus an honest, explicitly-open enumeration — not a better parser. This ticket is about
making the documentation self-checking, nothing else.

Context

• Parent: https://github.com/DIG-Network/dig_ecosystem/issues/2790 · epic
  https://github.com/DIG-Network/dig_ecosystem/issues/2760
• The guard and its docs: crates/dig-wallet/src/sage/sources.rs, statements at :26, :501, :508
• Five known mis-read shapes and their measurements are recorded in PR fix(wallet): make the provider registry real, and stop reporting a third party's peak #354's comment thread

Done condition

The worked examples execute in CI, and altering any expected value makes the test suite fail.

Effort

Low-medium. The examples exist and their true values are measured; the work is wiring them so they run.

[MichaelTaylor3d]

Lane claimed. Branch loop/360-367-batch, draft PR #368 (batched with #360 #361 #363 #364 #367). Base 4efa57f, v0.154.0 -> 0.155.0. Will comment the outcome here.

[MichaelTaylor3d]

Shipped on #368 (ca164a0)

The worked examples are now json blocks carrying their fixture and their expected sites /
ended_inside_a_test_item, and a test extracts every one from this file's own source and runs the
real classifier over it. Four examples, all executing.

Why not a rustdoc doctest -- this is the load-bearing decision

rustdoc does not run doctests on #[cfg(test)] items, and the whole sole_owner_tests module
is one. A fenced rust example there would compile in nobody's build while reading exactly like a
passing test -- the cfg-gated-and-unfalsifiable trap. It would have left the file in precisely the
state this ticket was filed about, while looking fixed. So the examples are executed by a harness
over include_str!("sources.rs"), which ties the assertions to the sentences a reader actually sees.

Fail-closed twice: an unterminated block is an error, and the example count is pinned -- so
deleting an inconvenient example, the cheapest route to green, fails rather than passing quietly.

No classifier heuristic was added

sweep and ends_a_column_0_item are byte-for-byte unchanged. Only the file's statements about
them are now enforced, as the ticket requires.

A second false claim, found by doing this

The trailing-comment example asserted sites=[5] and sites=[] for "the same fixture" -- and
that fixture was never written down, so its numbers were unreproducible by construction.
Reconstructing it: the natural 5-line shape yields sites=[5] for both terminators, because the
one-line fn later() { .. } itself ends at column 0 and clears the latch, so nothing is dropped and
the example demonstrates nothing. The shape that actually exhibits the defect needs a multi-line
production item, and it measures [6] and []. Both fixtures are now explicit in the docs.

A claim whose input is missing cannot be checked by anybody -- a worse failure than a claim that is
merely wrong, and the same root cause as the backwards example: an assertion nothing mechanically
contradicts.

Proof each example can fail

Five mutations, each run alone, each turning the suite red and naming the offending example:
altering any of the four documented sites values, and flipping a documented
ended_inside_a_test_item.

Sibling-lane note: this touches sources.rs only inside the #[cfg(test)] mod sole_owner_tests
doc comments plus one added test fn. It does not touch the sweep SCOPE or the anchored-root
resolver, which PR #369 owns.