fix(scan): align preview selection with scan execution

[Wsp030914]

Description

ocr scan --preview applied only static file filters, while a real scan also dropped files exceeding 80% of the resolved max_tokens value. This could make preview report a file as reviewable even though execution would skip it.

This change:

• resolves preview max_tokens using the same precedence as a real scan:
  CLI > saved config > template default;
• avoids initializing an LLM endpoint, preserving API-key-free preview;
• makes preview and execution consume the same scan-selection decisions;
• adds the stable too_large exclusion reason;
• preserves the real scan's existing skip logs and token-limit boundary.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change
• Refactoring (no functional changes)
• Documentation update
• CI / Build / Tooling

How Has This Been Tested?

• make test passes locally
• Manual testing

Additional verification:

• make check
• make coverage — 91.0%
• make build
• ocr review --audience agent — 0 findings across 5 selected production files
• integration coverage proving saved max_tokens affects scan preview
• token-limit boundary and no-limit unit tests

Checklist

• My code follows the project's coding style (go fmt, go vet)
• I have performed a self-review of my code
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally
• Documentation is unchanged because the existing CLI reference already states that scan preview reports per-file exclusion reasons
• I have signed the CLA

Scope

This PR only aligns deterministic scan file selection.

It does not attempt to predict:

• aggregate --max-tokens-budget exhaustion;
• resume reuse;
• LLM failures;
• other execution-dependent outcomes.

Related Issues

Closes #1040.

Related: #782 and #801 cover the corresponding diff-review preview behavior.

AI/LLM Disclosure

OpenAI Codex (GPT-5) assisted with implementation, test refinement, and drafting this PR description. OpenCodeReview using DeepSeek deepseek-v4-pro performed the final automated review. I manually reviewed and understand the complete final diff and take responsibility for the submitted changes.

[github-actions]

✅ OpenCodeReview: Review complete: 0 finding(s) across 5 selected item(s).