refactor(report): drop dead include_changed_files knob, fix config_rule stored_range docs

[cargo-affected-bot]

Two small accuracy fixes in the report layer, found by the nightly survey of docs/report-json.md (today's rotation bucket) read against src/report.rs.

config_rule reasons also have a null stored_range. The schema doc already says so — docs/report-json.md lists stored_range as "null for structural_backstop and config_rule" — but both Rust doc comments named only structural_backstop. selection.rs builds ConfigRule reasons with stored_range: None (a config rule matches an input path, not a coverage row), so the markdown was right and the doc comments on HitReason::stored_range and ReasonEntry::stored_range were incomplete. Corrected both to match.

SelectionInputs::include_changed_files has never been anything but true. It was introduced in #16 with two call sites (run and status), both hardcoded true; #71 folded those into the single plan.rs call site, still true. git log -S over the field's whole history shows no false anywhere. Its doc claimed "false collapses to selection.changed_files = None (no diff anchor was usable)" — but a run with no usable diff anchor is miss-no-reachable-sha, which routes to build_full_suite instead and never constructs SelectionInputs at all. So the branch described a state the type can't reach. Removed the field and the dead arm, leaving a comment on why the Some is unconditional. No behavior change.

Test. Neither change is a bug fix, so there's no failing-before regression test to add. What was missing is a pin on the encoding the corrected docs describe: selection.rs fills collect_sha, matched_hunk and stored_range for a ConfigRule reason by hand, and nothing asserted the resulting JSON. config_rule_reason_carries_no_coverage_anchor covers that — kind: "config_rule", stored_range: null, collect_sha: "", matched_hunk: [0, 0] — so a future edit that starts populating a range has to update the v1 schema doc deliberately rather than silently.

cargo clippy --all-targets, cargo fmt --check and the 113 unit tests pass locally. The functional suite runs in CI.

[cargo-affected-bot]

Both doc corrections check out against the code: selection.rs builds ConfigRule reasons with stored_range: None / collect_sha: String::new() / matched_hunk: (0, 0), and docs/report-json.md already documented the config_rule case, so the Rust comments were the stale side. The include_changed_files removal is safe — SelectionInputs has exactly one construction site (plan.rs::write_selection_report), which is only reached from the selection path in run.rs/status.rs; the no-anchor case goes through build_full_suite, which hardcodes changed_files: None.

One nit inline. cargo clippy --all-targets is clean and the 113 unit tests pass here (the description says 114 — that's the count including the test result: summary line).