shape-doctor.py: the doctor's git calls on a partial clone of the standard can lazy-fetch wherever a read is not already guarded; set GIT_NO_LAZY_FETCH=1 for the doctor's own git children so "the doctor fetches nothing" holds by construction #787

Goal: "the doctor fetches nothing" holds for every git command the doctor itself runs on a partial clone of the standard, by construction rather than by guarding each read. Only the NEXT lines it prints for a person may fetch.

Filed by lane openRepoShape-3 under v1.1.1, on Brett Heap's word of 2026-10-16T13:35:28Z ("yes create v1.1.1"). Lane 3's reviewer RV-f raised it as #754's N3, and again as N3 on #773 (#773 (comment)). #773 deferred it because it changes the environment of every row.

What is known:

• shape-doctor.py and update-shape.py: run from a partial clone of the standard, the shape currency row asks without fetching whether the clone holds what the comparison reads, and is CANNOT ANSWER with a fetch for a person to run where it does not, and a read that fails is refused, never a copy called both (#683) #742 and shape-doctor.py: the partial-clone row's NEXT fetches each commit it lacks objects at, by id, from the promisor remote whose branches hold it, so it no longer loops where the pinned commit sits only on another remote or another branch, and its why says where the clone does not hold the commit at all (#754) #773 guard the partial-clone path read by read. They ask rev-list --missing=print before any ls-tree, and never pass a commit the clone lacks to for-each-ref --contains, which on git 2.43 makes 1 fetch.
• RV-f measured that git 2.43 here honours GIT_NO_LAZY_FETCH=1. With it set, for-each-ref --contains prints "lazy fetching disabled" and makes 0 fetches. The variable dates from git 2.44 or earlier; check the first version.
• Any future read added to a row, in the doctor or in update-shape.py's Upstream as the doctor calls it, can lazy-fetch again unless it is guarded too.

What to do:

• Set GIT_NO_LAZY_FETCH=1 in the environment of every git child the doctor starts for its own reads: the doctor's git_out / git_answer helpers, and update-shape.py's reads where the doctor drives them.
• Do not set it in any line the doctor PRINTS for a person, since the NEXT fetch --refetch … must fetch.
• Check every row still answers the same on a full clone, and on a partial clone the same as today. A read that would now fail with "lazy fetching disabled" must become the existing CANNOT ANSWER row, not a traceback.
• Tests: a partial clone with a remote out of reach, where every row is asked and GIT_TRACE2_EVENT (or rev-list --missing=print before and after) shows no fetch.

Lands after #773 (#754).