fix(v1): compact long eval dashboard overrides

[hallerite]

Stacked on #2525 (codex/pin-subagent-timeout-semantics) to preserve the research evaluation base.

Change

Fold whitespace and abbreviate long override values in the native Rich evaluation overview only.
Long research prompts otherwise consume the viewport and hide the per-rollout rows. Full prompt
values remain unchanged in the evaluator and resolved configuration.

Validation

• Model-free diagnostic checked exact prompt preservation, serialization, defaults, nested skips,
  and 400-slot layouts at 80x24, 100x30, 120x40, and 160x48.
• Native live dashboard inspected at 120x40 and 160x48 with full 100x4 evaluations.
• 84 deterministic verifier tests passed; changed-file hooks passed.
• All-files hooks: markdownlint/format passed; lint reports an existing unrelated E402 in
  verifiers/v1/harnesses/hermes_agent/program.py:16.

Presentation-only change. No rollout budget, retry, prompt, or sampling change. Unsigned commit.

Note

Truncate and normalize whitespace in eval dashboard override display values

Adds _display_override in eval.py, which collapses whitespace runs to single spaces and truncates compact text longer than 60 characters to the first 40 characters plus an ellipsis and character count. Both override-segment construction paths (nested discriminator and ordinary field) now use _display_override instead of format_override. The underlying configuration and serialized representation are unchanged.

Behavioral Change: override strings shown by overrides now collapse whitespace and may be truncated; values previously displayed in full will appear shortened.

^{Macroscope summarized 773467e.}