Honor token boundaries in named search counts

[Widthdom]

Summary

Named-query search counts and summaries now forward --token-boundary to the reader. Queries such as CoverageObserver exclude longer identifiers, matching ordinary counts and named result rows while preserving count units and origin coverage.

The shared regression fixture covers isolated/repeated tokens, longer identifiers, zero matches, path and origin filters, exact/substring controls, human output, and counts above the presentation limit.

Fixes #5376

Validation

• Confirmed the new regression failed before the fix (expected 2, actual 6) and passes on net8.0 and net9.0.
• dotnet test CodeIndex.sln -c Release -m:1 -p:UseSharedCompilation=false: net8.0 12,612 passed / 9 skipped; net9.0 12,118 passed / 435 skipped; zero failures.
• Local Debug build and the issue's CLI reproduction: ordinary and named boundary counts both return 0.
• dotnet format CodeIndex.sln --verify-no-changes --no-restore --include src/CodeIndex/Cli/QueryCommandRunner.SearchRecipes.cs tests/CodeIndex.Tests/QueryCommandRunnerSearchTests.cs passed.
• Changelog fragment validation and git diff --check passed.
• Codex adversarial review of origin/main..HEAD: No blocking/actionable issues found.

Documentation

• Updated the English and Japanese sections of DEVELOPER_GUIDE.md and TESTING_GUIDE.md.
• Added changelog.d/unreleased/5376.fixed.md; CHANGELOG.md is unchanged.
• No agent instruction files changed. No follow-up issues identified.