Follow-up: reconsider bool / int return on Remove* methods (dime T4) #247

Origin

Dime Ultrareview cycle-1 on PR #241 (fix/device-validation-level-hardening), tracked finding T4 — surfaced by the improvement agent.

Finding

The four `Remove*(string id)` methods on `Device` and `Component` — `RemoveComponent`, `RemoveComposition`, `RemoveDataItem` — all return `void`. Callers cannot tell:

• whether the requested id was present in the tree at all (removal was a no-op),
• how many nodes matched (a duplicated id shows up as more than one removal),
• whether the tree shape actually changed.

For `MTConnectAgent.NormalizeDevice`'s Remove branch, this matters — the validation handler fires the `Invalid*Removed` event based on the assumption that the entity was present. If a race deleted the id concurrently, the event fires spuriously.

Rationale for defer

Changing the return type from `void` to `bool` or `int` is binary-API-breaking — every third-party consumer that has a Component-tree walker will see a signature change. It cannot land in the same PR as a bug fix that must be back-portable to point releases; it needs its own major-version cycle plus an explicit deprecation window.

Suggested phased design (for the follow-up PR)

1. Add new overloads: `bool TryRemoveComponent(string, out int removedCount)` etc., leaving the existing `void Remove*` in place.
2. Mark the `void` overloads `[Obsolete("Use TryRemove* overloads for removal reporting", error: false)]` in the next minor release.
3. In the following major release, remove the `void` overloads and promote `TryRemove*` to the canonical name.

Refs: dime cycle-1 T4 (improvement) — see PR #241's H1 recursion fix, which is the immediate trigger for wanting the removal count surfaced.