Add coverage summaries per production Bazel target #11

🎯 Motivation

The global coverage result is too coarse to identify weak production components. The file-level HTML report is too detailed and requires consumers to aggregate files manually.

LCOV is only raw coverage data and does not contain file/line justifications. It therefore cannot represent the effective coverage status and is unsuitable as the consumer-facing target-level result; it should remain available as a low-level diagnostic.

A production-target view would make coverage status easier to assess, communicate, compare, and integrate into dashboards or release evidence.

💡 Proposal

Add coverage summaries per production Bazel target, based on the union of all relevant tests and including the effect of applied justifications.

🖥️ Human-readable

Provide an HTML overview with:

• target label and understandable status;
• raw/effective line and branch coverage;
• covered, uncovered, and justified counts;
• links to the existing file-level details.

🤖 Machine-readable

Provide a stable, self-contained format with the same target-level information, including raw/effective counts and percentages, plus enough justification information to explain the effective values. Existing global and low-level file outputs remain available.

❓ Open semantics

• Direct target sources only, or transitive production dependencies?
• How are files shared by multiple targets assigned or counted?
• How are declaration-only, no-data, and similar files handled?
• One schema across all supported coverage backends.

✅ Acceptance criteria

• Target-level effective coverage is available in HTML and machine-readable form.
• Coverage from all tests is merged before aggregation.
• Justifications are reflected and explainable in the effective values.
• HTML links to file-level details.
• Shared-file and dependency semantics are documented and tested.
• LCOV remains available as raw low-level coverage data, but is not required to encode justifications.