Cardano Updates

Also fixes multiple labels

The cost-dashboard is now just 3 static files (index.html, controller.js,
view.css) with no build step. The controller.js uses an IIFE pattern to
expose the Controller global, eliminating the need for webpack bundling.

Removed:
- CostModel.agda, Main.agda (outdated Short Leios model)
- webpack.config.js, package.json, package-lock.json (build infra)
- build.sh, debug.sh (nix-shell scripts)
- shell.nix symlink, .envrc, .gitignore

Updated site/static/cost-estimator/ and site docs to match.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Replace the old Short Leios model (IBs, pipelines, phases) with the
Linear Leios (CIP-164) cost model as documented in docs/cost-estimate/.

Key changes:
- Remove Input Blocks (IBs); EBs now carry 32-byte tx hash references
- Main input is confirmed throughput (TxkB/s) instead of tx/s
- Add P(cert) = (1-f)^voting_window certification probability
- CPU model: Apply + Reapply + EB header + Vote + Cert validation
  (matches doc values exactly: 97.1 ms/s at 5 TxkB/s)
- Egress model: per-component with network topology parameters
  (inbound peers, fetch multiplicity M, vote spanning-tree)
- Storage: tx closure + EB body (P(cert) cancels) + EB headers + RB
- IOPS: UTxO-HD model with tx data + UTxO state updates
- All default values from CIP-164 simulation config and benchmarks
- Add CPU breakdown display (Apply, Reapply, EB, Vote, Cert)

Cross-checked against cost-estimate doc tables at all throughput levels.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

This includes redundancy on fetched txs and votes (from up to 2 peers)

Along with a lot more changes on network egress

This allows for a better comparison and stresses that linear leios does
not result in more costs when load is < Praos capacity.

The worst-case numbers were unrealistic (plutus block limits prevent all
txs to have full plutus tx limits)