policy: dump_usb_ids must remain (tier-1 source for online-data aggregation)

[zackees]

Policy: keep crates/fbuild-core/examples/dump_usb_ids.rs

Recording an explicit anti-removal policy after coming close to dropping
this in #719.

Why it must stay

dump_usb_ids runs inside the Update data workflow (.github/workflows/update-data.yml)
as tier-1 of the USB-VID name merge. It exports the bundled
usb-ids Rust crate's snapshot as
JSON, which tools/merge_sources.py then unions with linux-usb.org,
the usbids GitHub mirror, and the curated vendor_names_inlined.py
overlay to produce the published online-data/data/usb-vid.json.

Removing the example would:

• Drop a fourth independent source of truth from the aggregator (we
  go from 4 → 3, losing whatever the usb-ids crate carries that the
  text mirrors don't).
• Force the usb-ids workspace dep to go too (since nothing else uses
  it), losing the ability to cross-check the curated overlay against
  the upstream Rust snapshot.

Decoupling that's already in place (#719)

The fbuild runtime USB resolver (fbuild-core::usb::resolver) no
longer reaches for the usb-ids crate at all — it uses the compile-time-
embedded usb-vendors.tar.zst (fbuild-core::usb::embedded). The
usb-ids dependency is now only pulled by the dump_usb_ids example,
which is built solely by the nightly workflow.

Action

This is a documentation-only issue. Add a banner comment to
crates/fbuild-core/examples/dump_usb_ids.rs referencing this issue
(so future LLM-driven cleanup passes don't try to remove it), and
mention it in crates/CLAUDE.md under "Crate Responsibilities".

No code changes required beyond the banner.