Track interpreter code size in CI

[kadircanyildirm-crypto]

Nothing in CI measures the size of the interpreter: the only match for "size" under .github/ is a comment about usize. Meanwhile #187 asks for SIMD to be feature-gated "to allow lower code size for deployments that don't need this", so size is a requirement that no job checks, and a regression would only be noticed by hand.

Proposal: a size job that builds spacewasm in release for an agreed set of targets and records .text/.rodata/.data (llvm-size over the archive, or cargo-bloat) into size-results.json next to benchmark-results.json, so the existing readResults/upsertComment plumbing in comment.yml can post a "Code Size" comment with the delta against main, the same way the CoreMark comment works.

Three questions before writing anything:

1. Targets: host and i686, or the bare-metal ones CI Benchmarking #178 is bringing in (riscv32imc, thumbv7em)? If the latter, this should wait for CI Benchmarking #178 and reuse its builds rather than add another cross setup.
2. Feature sets: default only, or one row per optional feature once SIMD is gated?
3. Report only, or a threshold that fails the job?

Happy to implement it in .github/ once the shape is agreed.

AI use per AI_POLICY.md: Claude Code assisted with surveying the workflows and drafting this issue.