Verifies that every fork with an embedded NUT bundle produces an
activation block containing exactly the bundle's deposit transactions.
Discovery uses forks.All + derive.UpgradeTransactions, so future forks
with NUT bundles are covered automatically without test changes.

This reverts commit 941e15a.

The kona-hardforks build.rs walks ancestors of CARGO_MANIFEST_DIR to find
op-core/nuts/bundles/ during compilation. The kona-{client,host,node}
images scope their Docker context to rust/, so the ancestor walk can't
reach op-core/ and the images fail to build.

Pass op-core/nuts/bundles as an additional named BuildKit context
(nuts-bundles) and copy it into /workspace/op-core/nuts/bundles so the
ancestor walk succeeds. Keeps the primary rust/ context small and avoids
mirroring the bundle JSON into the crate tree.

The Hardfork impl for each NUT-bundle-backed fork is identical boilerplate
(decode, EIP-2718 encode, sum gas). Move that pattern into a single
internal macro so adding future forks is a one-liner next to the
build-script-generated constructor.

Split parsing + codegen into build_helpers.rs, included via #[path] from
both the build script and a new integration test. Use anyhow to carry
error context up to a single panic in main. The integration test runs
the generator on a fixture JSON and asserts the exact Rust source output,
so codegen changes surface as a test failure rather than a downstream
compile error.

Follows the per-fork test convention from ecotone/fjord/isthmus/holocene.
Asserts the EIP-1967 implementation slot of representative predeploy
proxies (L1Block, GasPriceOracle) changes across Karst activation — a
Karst-specific smoke test that the bundle's proxy upgrades took effect.

Not generalized into the NUT bundle activation test because it does not
hold for future forks that may not upgrade proxies.

The generic activation test was missing the core RunFaultProofProgram
call (preceded by BatchMineAndSync to advance the safe head), so it
only exercised the sequencer, not the proof pipeline it lives under
actions/proofs/ to validate.

Fold the Karst-specific impl-slot assertions back in via a switch on
fork name, establishing the pattern for future forks to register their
own post-activation checks alongside the universal ones. Delete
karst_fork_test.go.

Generated bundle constructors now wrap construction in a function-scoped
once_cell::sync::OnceCell and return &'static NutBundle. The first call
allocates and stores the bundle; subsequent calls return a cached
reference. Activation-path calls to txs() and upgrade_gas() no longer
reallocate on each invocation.

Consumers are unchanged — method calls auto-deref through the reference.

cargo fmt -- collapses a few multi-line statements onto one line per
the use_small_heuristics = "Max" rustfmt.toml setting.

cargo clippy -- replaces a match-on-Option in the bundle codegen with
Option::map_or_else per the option_if_let_else lint.

The cannon-repro.dockerfile's client-build stage runs cargo build on the
kona workspace, which triggers the kona-hardforks build script. That
script walks ancestors of CARGO_MANIFEST_DIR looking for op-core/ and
panicked because the context only contained kona/. Pull in
op-core/nuts/bundles from the existing 'monorepo' named context so the
ancestor walk succeeds — same pattern as the kona-node/host/client
image fix.

sync::OnceCell with the critical-section feature requires an impl of
_critical_section_1_0_{acquire,release} to be linked in, which the
MIPS64 cannon and RISC-V asterisc fault-proof VM targets don't provide.
This broke kona-client on those targets.

Switch the generator to emit once_cell::race::OnceBox, which uses a
single AtomicPtr for the cache slot — lock-free and requires no
critical-section impl. The NutBundle is heap-allocated via Box on first
call and then cached as &'static NutBundle for all subsequent calls.
Enables the `race` feature on once_cell.