test(codec): ignore the zstd encoder tests on wasm32 rather than abort the suite - #389

What

Gate the eight zstd encoder tests in nodedb-codec/src/zstd_codec.rs with
#[cfg_attr(target_arch = "wasm32", ignore = "...")]. Test attributes and
comments only; no behaviour change — the wasm arm still returns the same error.

Why

nodedb-codec already splits the codec by target: C libzstd off wasm, ruzstd on
it (nodedb-codec/Cargo.toml:31-36), and compress_native's wasm arm is an
explicit stub returning CompressFailed (nodedb-codec/src/zstd_codec.rs:312-320);
no fallback sits under it, and pipeline propagates that error for
ColumnCodec::Zstd. The eight tests call encode/encode_with_level, so they panic —
and on wasm32-wasip1 a panic aborts the process: the suite dies at the first
one and reports nothing after it. That makes the crate's wasip1 suite unusable,
which is what a shared-crate wasm job for NodeDB-Lite needs.

That stub is a design choice, not a missing dependency: ruzstd 0.9 ships an
encoder (ruzstd::encoding — pub mod encoding in lib.rs, compress in
encoding/mod.rs, FrameCompressor defined in encoding/frame_compressor.rs and
re-exported from encoding/mod.rs) that this crate has not adopted, and the C
library would need a C-to-WASM toolchain. Adopting either one changes what the wasm
build does, so it is deliberately not part of this change. The stub's own comments
used to call it a fallback that encodes "a minimal Zstd frame" — it never did, and
they now describe the stub.

cfg_attr(ignore) keeps them compiled and listed by --list on that target, so
the gap shows up in the run. cfg(not(target_arch = "wasm32")) would delete them
from the target instead, and a --skip list in a consumer workflow would record
the gap as a number in someone else's file.

How to check it

# native: the eight still run, and nothing is ignored
cargo test -p nodedb-codec
# 263 passed; 0 failed; 0 ignored

# wasip1: the four aborts owned by other changes are skipped, to reach this one
RUSTFLAGS="" CARGO_TARGET_WASM32_WASIP1_RUNNER="wasmtime -W max-wasm-stack=33554432 --dir=." \
  cargo test --profile ci --target wasm32-wasip1 -p nodedb-codec -- \
  --skip delta::tests::rejects_overflowed_noncanonical_and_huge_input_before_allocation \
  --skip double_delta::tests::huge_count_and_noncanonical_tail_are_rejected_before_decode_work \
  --skip fastlanes::codec::tests::hostile_counts_and_block_shapes_fail_before_allocation_or_looping \
  --skip vector_quant::opq::tests::top1_recall_on_training_set
# 252 passed; 0 failed; 8 ignored; 4 filtered out

Red arm: with the ignores removed and the same four skips, the same command
aborts —

test zstd_codec::tests::better_ratio_than_lz4 ... Error: failed to run main module
exit 134

Each ignored test names its reason:

test zstd_codec::tests::empty_data ... ignored, nodedb-codec has no wasm encoder: encode() returns CompressFailed

cargo fmt --all -- --check exits 0; cargo clippy -p nodedb-codec --all-targets -- -D warnings exits 0 natively and for --target wasm32-wasip1.

Scope

• The module's other four tests stay live on both targets: three are decode-side,
  and streaming_input_limit_precedes_buffer_growth drives the encoder's
  declared-limit check, which rejects before any encoder work.
• The four --skip flags above name tests owned elsewhere: three abort through a
  32-bit overflow (delta::tests::rejects_overflowed_noncanonical_and_huge_input_before_allocation,
  double_delta::tests::huge_count_and_noncanonical_tail_are_rejected_before_decode_work,
  fastlanes::codec::tests::hostile_counts_and_block_shapes_fail_before_allocation_or_looping)
  and are fixed by an open change of their own;
  vector_quant::opq::tests::top1_recall_on_training_set is a recall threshold
  the target misses and stays a named skip in the consumer job, not here.
• No CI workflow is added here; this is the prerequisite for the shared-crate
  wasm job that belongs in the NodeDB-Lite repository.