Changes touching this path

• a push against a forge head it does not hold refuses before any upload, and a long-running verb no longer writes an older position over a newer one: stow_batches reads the heads of a forge on /ref with the failure kept, where an unanswered /ref or a forge head this position does not hold read as a have of nothing and stowed the whole history into the nginx 413 of the PR 2531 land, and from a complete position it refuses a push past a forge head it neither holds nor retires with the first line of the forge refusal, which /ingest still makes; DagRepo::save_to and Workspace::persist write the heads, working change, conflict set, settled ledger and draft slots only when the handle changed them since it read them, and refuse when another command moved them in between, where every save wrote them from memory and a loot pull-grants open for forty minutes put the head back under a land adopt; pull-grants files its keys and saves once, before the ack, where it persisted once a grant; a save records what it wrote rather than reading the position back, so store_file_reads stays at 24 on the gated workload, where a read-back after each write took it to 28 and the land was refused. pinned: a_push_against_a_forge_head_it_does_not_hold_refuses_before_any_stow over a proxy logging each request, a_forge_whose_ref_cannot_be_read_refuses_the_push, a_pull_grants_opened_before_the_position_moved_does_not_move_it_back, a_save_that_did_not_move_the_position_leaves_a_newer_one_alone and a_save_that_moved_the_position_refuses_over_a_newer_one. red under named mutations, each restored: the forge heads swallowed as before (2 failed and 1 passed over the three sync tests), no unseen-head refusal (1 and 2), the old save that always writes (2 failed and 0 passed over the two engine tests), every position read as changed (1 and 1 over the engine tests, 1 and 2 over the sync tests), no moved check (1 and 1), the draft slots read as changed (1 and 2), the working change left out of the compare, so a finalize was not saved (2 failed and 1 passed over two deposit tests and a snapshot test), and an empty heads set held apart from no heads file (3 failed and 0 passed over a mergetool, a relay push and a dock-files test). a push to a namespace of another key is now refused at /ref, before /wants could name the ownership rule. cargo test green, 4876 passed over 144 test result lines with 15 ignored, and the site gate green at 915 passed. the change lane is still not bounded by the batch budget, so a first push of a history larger than the body cap to an empty forge still fails, left for a follow-up. the perf gate reads no move. owes no forge deploy; lands use the push refusal once the release binaries of the primary are rebuilt; it also files kqzkruyv, the unbounded change lane, and kzknpprm, the phantom deletions of folded files (kkpruryt) 868cea49 · dbf3dbe6…

Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.