Promote Develop: Read state= in drive-pr, Fetch the Prose Gate From One Revision, and Place Copilot's Outstanding Findings - #3703

…in hub_fetch_run.py (#3696)

## Summary

When `resolve_main_commit()` cannot resolve a SHA, `hub_fetch_run.py`
fetched the prose gate's companion and the script from moving `main` as
separate raw URLs, so a push between the requests could pair revisions.
In that case the helper now fetches the companion and script sequence
twice and refuses (exit 1, "the gate did not run") when any file differs
between the two reads. A resolved commit is still fetched once.

## Why this design

The raw-only hosts that reach the unresolved path have no API, git, or
archive route, so a single immutable fetch is unavailable there.
Refusing outright would break the documented raw-only fallback. A
bracketed double read needs no new endpoint, stays standard-library
only, and turns a push landing mid-run into a loud refusal that names
the re-run.

Residual limit, found in review and deliberately stated:
raw.githubusercontent.com caches each URL separately at its CDN, so
right after a push the two URLs can serve different revisions
consistently, and the bracket cannot see that. The docstring claims only
the mid-run push case. Closing the CDN case needs an immutable fetch (an
archive, or a version stamp shared by script and data), which is a
larger design change than this issue's snippet-only scope.

Docs in GOVERNANCE.md and the husky/pre-commit READMEs describe the
unresolved fallback as still running the gate; they do not mention the
new refusal. They were out of this change's file claim and are reported
for a follow-up.

Closes on promotion: #3533

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>

…t Only Lineless Files as Covering (#3702)

## Summary

- **#3672, a Copilot findings list it noted rather than placed.**
- `scripts/reviewers/copilot.json` gains the `outstanding findings:` and
`unresolved review findings:` spellings under `labels` and
`finding_lists`. This is a vocabulary change with no new code branch.
- The reader now reads numbered findings lists as well as bulleted ones.
Each item keeps its text and the `path:line` ranges its code spans name.
- A span counts only where its path is a changed file or a thread's
path, so `node:20` is not read as a line.
- FINDINGS WITH NO THREAD prints every item with a mark: `[thread at its
line]`, `[no thread at its line]`, `[outdated thread on its path]`,
`[thread page cut, unknown]`, or `[no path:line code span]`.
- An item whose span no thread sits on now counts toward that block's
number, even where the open count and the linked entries balance.
- **#3489, a file table leaving out only binary files.**
- The changed-file query now reads `additions`, `deletions`, and
`changeType`.
- A file added, modified, or deleted with no added or deleted line may
go unnamed in the table, which then reads `coverage=table`. A rename, a
copy, a type change, and a file missing a count are never left out.
- Nothing is left out unless a file with an added or deleted line
remains. The reason line says how many files were left out.
- **#3701, a current file-level thread read as outdated.** Copilot's
first round raised it on this branch. `Q_FULL` now reads each thread's
`isOutdated` flag, so a null `line` no longer stands for outdated. A
file-level thread marks no item, and the item stays counted.
- `scripts/pr_review.py --help` and `scripts/README.md` describe these
changes.

## How a Binary File Is Known

GitHub reports `additions=0` and `deletions=0` for a binary file. Its
changed-file list carries no binary flag, and neither the GraphQL nor
the REST shape adds one.

A constructed case: a pull request changes `docs/a.md` (3 added lines)
and adds `static/b.png` (0/0). Copilot's table names `docs/a.md` alone.
The digest now reads `coverage=table` with the reason "names exactly the
1 changed file(s), 1 more with no changed line left out". Before this
change it read `coverage=none`.

Two alternatives were not taken. Guessing from the extension misses an
unusual binary and wrongly excuses a text file with a binary-looking
name. Reading `.gitattributes` would mean a content fetch per pull
request.

## New Exit Path

A findings-list label followed by no list the reader can read now leaves
a note rather than being placed silently. An example is `Unresolved
findings:` followed by `None.`. At the top level of a body, that note is
also an unplaced part, so `status` exits `50` until its fingerprint is
acknowledged with `comment --fingerprint`. Inside a collapsed section it
is a note only.

## Remaining Findings

- **The maintainer's call on the #3489 trade.** An empty file, a
mode-only change such as dropping an executable bit, and a text file
that `.gitattributes` marks `binary` or `-diff` all read as 0/0.
GitHub's file list does not tell them apart from a binary file. Each one
therefore may go unnamed in a Copilot table, and the table still reads
`coverage=table`. No reviewer or diff showed the `.gitattributes` case.
The README names this residual. Whether excusing these is acceptable is
open.

## Local Review Findings

Four local strict-review passes ran. The first two raised findings that
were fixed on this branch. The third raised the #3701 finding, which is
fixed here after Copilot raised it too. The fourth, on the fix, raised
none.

Closes on promotion: #3672, #3489, #3701

🤖 Generated with [Claude Code](https://claude.com/claude-code)