upstream: asic-rs (crates.io)
version: 0.8.1 (cargo-vendored under vendor/, per ADR 0003)
state: pristine cargo-vendored tree + this patch queue.

asic-rs is materialized with `cargo vendor vendor` and consumed through
.cargo/config.toml's vendored-sources replacement (directory = "vendor").
Unlike wx-rs, asic-rs is many crates (asic-rs, asic-rs-core, and the
per-make / per-firmware crates); a change to any of them lives here as a
numbered .patch, NOT as an in-place edit of the pristine tree (ADR 0003).

The committed vendor/ tree already has the patches applied AND its affected
crates' .cargo-checksum.json refreshed, so a normal build needs nothing extra
-- the patch queue is provenance + the re-vendoring recipe.

GOTCHA (cargo caching): cargo fingerprints a registry/vendored crate by
(name, version, source), NOT by file content, so it will REUSE a cached
build of a vendored crate even after its source is edited. CI is unaffected
(fresh target/ each run compiles the patched source). But an incremental
LOCAL build that already compiled the unpatched crate must be forced:
  cargo clean -p asic-rs-firmwares-whatsminer   # then rebuild
Otherwise the local binary silently keeps the old behaviour.

Update / re-vendor flow:
  1. bump the pin in crates/alkane-gui/Cargo.toml, `cargo vendor vendor`
     (this regenerates a pristine vendor/ and DROPS our edits)
  2. scripts/apply-asic-rs-patches.sh   # re-applies patches/ + refreshes checksums
  3. resolve any patch that no longer applies, rebuild, commit

patches/:
  0001  whatsminer set_power_limit (round-9 #4): fix the v3 (WhatsMinerOS)
        privileged auth -- the "super" account's token password is the separate
        v3 credential (default "super"), not the web/admin password -- and send
        the watts as a bare integer; also treat the apply-restart's transient
        RPC error as success. See the patch header.
  0002  antminer Equihash (Z15/Z15J) hashrate (round-10 #8): the cgminer "GHS"
        key carries a Sol/s value, so use the base Hash unit for Equihash so the
        SI prefix is correct ("288k", not "288T"). See the patch header.
