Distribute generated RTL without deploy-branches (regen / RTL repo / release assets)

[DanielKellerM]

Current mechanism

util/deploy.py distributes generated RTL by, for each release, temporarily un-gitignoring target/rtl, git add .-ing the generated RTL, and pushing a __deploy__<hash>__<branch> branch (the two-tag scheme: vX.Y.Z-src source / vX.Y.Z deploy). Downstreams Bender-depend on the deploy branch/tag because source branches gitignore target/rtl and Bender needs a concrete RTL flist.

Why change it

• The __deploy__* branches proliferate on the repo and the GitLab mirror.
• Deploy branches generate a representative configuration, but most of the cases users will have to modify the make file for their own protocols, generating dirty uncommited local files.

Options to weigh

1. Downstream regeneration — consumers run make idma_hw_all themselves; nothing generated is stored. Purest (single source of truth = templates + src/db), but pushes the Python/MARIO toolchain (peakrdl, mako, bender-pickle) onto every downstream build, and Bender has no native pre-generate hook.
2. Separate generated-RTL repo — a thin iDMA-rtl repo holding only the generated RTL, auto-pushed by CI on release; downstreams Bender-depend on it. Keeps Bender's git-dep model, separates generated from source, removes __deploy__* from the main repo. Cost: a second repo to maintain.
3. GitHub Release assets — CI attaches the generated RTL as a release tarball; downstreams fetch + unpack. Versioned + durable, but Bender fetches git repos, not release assets, so it needs a fetch/vendor step.
4. CI artifacts (actions/upload-artifact) — cheapest to produce but expire (~90d) and aren't cleanly versioned — CI-internal only, not for downstream consumption.