Tail-form return in terminal switches

[AaronWebster]

When the optimized Ok() switch is the function's tail — last group emitted, no [requires] clause after — arms that validate exactly one field with no residual collapse from

case K:
  if (!field().Ok()) return false;
  break;

to

case K:
  return field().Ok();

Saves one conditional branch per qualifying arm. On Thumb-2 / MicroBlaze (where -Os is conservative about merging cmp+beq sequences) and on older compilers that don't see through the if/return-false/break shape, this trims a few bytes per case across large tagged-union schemas.

Per-arm decision: any arm that doesn't qualify (multi-field, residual, or the switch isn't terminal) keeps the break-form body. Mixing forms within one switch is safe — the default fallthrough lands on the function's return true; either way.

Golden churn: many_conditionals.emb.h shrinks substantially (the 99 single-field cases all collapse to tail-form); condition.emb.h and parameters.emb.h see smaller similar collapses.

Size impact (cumulative vs. master)

Target	Metric	Master	PR	Delta
ARM Thumb-2	TU .text	18962	17974	−988 (−5.2%)
ARM Thumb-2	LargeConditionals::Ok()	5382	4930	−452 (−8.4%)
ARM Thumb-2	DisjunctionConditionals::Ok()	358	262	−96 (−26.8%)
MicroBlaze	TU .text	43640	35796	−7844 (−18.0%)
MicroBlaze	LargeConditionals::Ok()	14824	8308	−6516 (−44.0%)
MicroBlaze	DisjunctionConditionals::Ok()	640	532	−108 (−16.9%)
Host x86-64	TU .text	29166	28242	−924 (−3.2%)
Host x86-64	LargeConditionals::Ok()	3948	3040	−908 (−23.0%)
Host x86-64	DisjunctionConditionals::Ok()	431	354	−77 (−17.9%)

Note: LargeConditionals::Ok() measured by nm grows slightly on Thumb-2 and MicroBlaze for this PR alone (+302 / +604 bytes), but TU .text still shrinks. The compiler is inlining tail-form arms into callers more aggressively, so reduced overall code lives in Ok()'s callers rather than Ok() itself.

Stacked on #257.