Polish: InvariantViolated reason text + BoundsUnit per-variant Display pin

[r6e]

Source

PR #144 R1 review — bundled low-priority polish items.

Items

1. InvariantViolated.reason issue-ref fragility (code-reviewer SUG-2 conf 80)

The unreachable!() replacement at pak/mod.rs for the stream_entry_to dispatch arms uses:

reason: "stream_entry_to dispatch reached an unsupported \
         CompressionMethod arm — early-reject at top of \
         function was bypassed (see issue #138)"

The (see issue #138) reference is fragile: if #138 is closed and a new refactor reintroduces the bug, operators following the link land on the closed unreachable! removal ticket, not on whatever new issue tracks the regression.

Fix: drop the issue reference (the message text is self-explanatory and git-blame is the source of truth) OR rephrase to flag the staleness explicitly: "… was bypassed; #138 introduced the typed fallback — file a new issue if this fires".

File: crates/paksmith-core/src/container/pak/mod.rs (the InvariantViolated reason inside the match arm at the bottom of stream_entry_to).

2. BoundsUnit per-variant Display pin (pr-test sev 5)

BoundsUnit::Bytes => "bytes" and BoundsUnit::Items => "items" are wire-stable strings consumed by the same dashboards as the WireField/AllocationContext tokens. They're transitively exercised by IndexParseFault Display tests but with uneven coverage. PR #144 added per-variant pin tests for the two new enums; BoundsUnit should get the same treatment for consistency.

Fix: add bounds_unit_display_tokens_are_wire_stable modelled on the new wire_field_display_tokens_are_wire_stable (~5 lines).

File: crates/paksmith-core/src/error.rs (next to the WireField/AllocationContext pin tests).