Fix unsound Ok() switch guard by demoting all-residual conditional discriminants

[AaronWebster]

Problem

The optimized Ok() switch (introduced in #253, extended in #256/#257) reads its
discriminant once and emits, as a safety net:

const auto emboss_reserved_switch_discrim = <discriminant>;
if (!emboss_reserved_switch_discrim.Known()) return false;
switch (emboss_reserved_switch_discrim.ValueOrDefault()) { ... }

When the discriminant is itself a conditionally-present field read from a buffer too
short to contain it, it reads as Unknown, so this guard makes Ok() return false —
even though the message is valid, because the fields gated on the (absent) discriminant
simply don't exist. This is latent on master: no existing test schema produces the
shape.

Same bug that #266 fixes, reached by a simpler route.

Fix

Instead of #266's dedicated ok_method_switch_block_guarded template, add one rule to
the demotion pass in _generate_optimized_ok_method_body: a switch group that is not
provably-known and has no bare arm (every entry carries a residual) is demoted to
per-field has_X() checks — the existing ok_method_test path — instead of being
emitted as a guarded switch. After this, _emit_switch_block only ever sees
provably-known or bare-arm-guarded forms, so the unsound guarded-switch template is
deleted outright.

Soundness is observationally identical to #266's guarded fallback: with the discriminant
Unknown, a statically-false residual makes has_X() Known-false (field absent → valid,
the exact case #266 fixed), otherwise has_X() is Unknown → Ok() returns false; with
the discriminant Known it computes the full has_X(), same as a dispatched arm minus
the pointless switch.

Verification

• Proven byte-identical generated code on every existing schema; the golden diff is
  confined to testdata/golden_cpp/condition.emb.h.
• Keeps Guard conditional switch discriminants in optimized Ok() #266's regression tests (testdata/condition.emb +
  compiler/back_end/cpp/testcode/condition_test.cc); assertions unchanged, comments
  reworded per review.
• Full bazel test //... green (206/206).
• Size: ~0 delta on all targets (MicroBlaze LE / Vitis authoritative) — this is a
  correctness fix, not a size optimization.

Supersedes

Supersedes #266 (same bug, simpler fix; #266's regression tests are carried over here).

[robrussell]

This last phrase is unclear
"(A if (has_tag().ValueOrDefault()) wrapper would wrongly accept this message.)"

Does that reflect the state before this PR or after? Is that intended behaviour? Some of the explanations in the comments for these tests might be helpful in the language docs (either as clarifications or examples).