fix: clear the warning-sink ThreadLocal and drop dead test state - #146

Fixes the 5 Sonar issues still open on master after the setAccessible ones were accepted. 263 tests green, branch coverage unchanged at 303/303.

The real one

TypeConverter:31 — Call "remove()" on "ACTIVE"

ACTIVE was a ThreadLocal created with withInitial(...) and only ever set, never removed. YamlFileInterface.load(Object) installs the plugin's logger and restores the previous value in a finally — but "restoring" wrote the default back rather than clearing the entry, so every thread that ever loaded a config kept a sink referenced for its lifetime. On a server thread pool those threads outlive the load.

It now starts unset, warn() falls back to LOG when nothing is installed, and restoring a null sink calls remove():

static @Nullable Consumer<String> pushSink(final @Nullable Consumer<String> sink) {
    Consumer<String> prev = ACTIVE.get();
    if (sink == null) {
        ACTIVE.remove();
    } else {
        ACTIVE.set(sink);
    }
    return prev;
}

The existing finally { pushSink(prev) } call site needed no change — prev is now null on the first install, which clears the entry.

Dead state removed

• LeniencyTest:7 — unused java.nio.file.Files import.
• NamespacedKey:7 — the ns field was assigned and never read. The double exists so tryFormatAsKeyed can call key() then value(); the namespace was never part of that. Field and constructor parameter dropped, and the single call site in SoundImpl updated.
• Sounds:9 — same story. The tests only compare identity (Sounds.MY_SOUND_ROCKS), so the name field, the getSound factory and the @NotNull import all went. The class deliberately keeps no public String constructor and no toString() override — that is what makes YamlWriter fall through to static-field matching — so the comment now records it.

Kept on purpose

StaticFieldTestClass:11 — PRIVATE_INSTANCE

This one is a false positive worth explaining rather than obeying. The field is read only by reflection, from YamlWriter#getStaticFieldName, and it is the only case exercising the field.canAccess(null) == false branch.

Measured rather than assumed — deleting it:

getStaticFieldName BRANCH  covered=7 missed=1     (was 8/8)

So it stays, with @SuppressWarnings("unused") and a comment recording why, which is also what stops the issue reappearing.

On the coverage numbers

A local build shows 2 uncovered lines in YamlWrapperFactory (the v1-vs-v2 selection). That is not a regression and not related to this change: only one SnakeYAML version can take that branch per run. Verified the union across report-1.14.xml and report-2.4.xml leaves no uncovered lines, which is why Sonar — aggregating all four reports — reports 100%.