Implement template system steps 1-3

Scope: implement only steps 1, 2, and 3 of the template-system roadmap.

1. Expose templates through the CLI

• Add --wordlist-templates for selecting bundled template sources.
• Add repeatable --template-var KEY=VALUE overrides.
• Add --template-strict to reject unresolved or invalid placeholders.
• Add --template-report FILE for a machine-readable expansion report.
• Preserve existing -w and --wordlist-categories behavior.
• Document the flags and add CLI/config parsing tests.

2. Compile a deterministic template plan

• Add dependency-light TemplatePlan, TemplateContext, and TemplateSource data classes.
• Add compile_template_plan(...) as the validation and planning entry point.
• Validate placeholders, variables, and referenced categories before the scan starts.
• Freeze dynamic values such as dates once per plan.
• Estimate expansion cardinality and enforce a configurable budget.
• Make deduplication, ordering, and shard assignment deterministic.
• Preserve source and variable provenance and expose stable to_dict() output for Python and JSON consumers.
• Test strict mode, invalid inputs, budgets, determinism, and serialization.

3. Select templates from preflight evidence

• Add a pure TechnologyMatcher that converts fingerprint/preflight evidence into explicit technology matches.
• Add a pure TemplateSelector that maps matches to template sources.
• Support --wordlist-templates auto while exposing selected sources.
• Keep selection evidence-based and inspectable; avoid opaque numeric scoring.
• Define a deterministic fallback for missing or ambiguous evidence.
• Include selected templates and supporting evidence in the template report.
• Test known technologies, mixed evidence, ambiguous evidence, and the fallback path.

Completion criteria

• Identical inputs produce identical plans and candidate ordering.
• Python API and CLI expose equivalent structured planning output.
• Template planning/report failures do not corrupt scan state.
• Existing wordlist workflows keep passing their regression tests.