feat(engine)!: retune the template publish fee curve

[sdbondi]

Motivation

Publishing a mid-sized template costs around 5 tTARI, which is high for the common case.

Measuring the sizes that are actually published — the scaffolded release profile, followed by the wallet daemon's -Os pass (applications/tari_walletd/src/services/wasm_optimizer.rs:43) — gives this corpus:

template	published size
bare #[template], no Component	25 KiB
hello_world	76 KiB
tariswap	136 KiB
basic_nft	151 KiB
stealth	198 KiB
access_rules	260 KiB

So the 5 tTARI case is a large-but-ordinary template at ~260 KiB, which the 30 KiB free allowance prices at 5.14 tTARI.

Change

Free allowance 30 KiB → 96 KiB. This brings 260 KiB to ~2.7 tTARI while leaving the tail where it was: at 512 KiB and beyond, 66 KiB of extra allowance is a rounding error against the excess, so oversized templates stay expensive.

New flat per_template_publish_cost, 250,000 µT. Raising the allowance alone drops a minimal template to 0.10 tTARI. That is a structural limit of a size-only model, not a tuning miss: a published binary carries ~25 KiB of fixed template_lib + CBOR machinery before any author code, and a minimal component template lands around 76 KiB — so any allowance that covers library overhead also makes a trivial publish nearly free. The flat charge separates the two concerns: the allowance decides where the premium starts to bite, the flat charge decides what the cheapest publish costs. It folds into the existing FeeSource::TemplatePublish breakdown entry, so no new fee source.

Resulting curve

published	before	after
minimal 25 KiB	0.03 T	0.28 T
hello_world 76 KiB	0.25 T	0.35 T
tariswap 136 KiB	1.12 T	0.52 T
basic_nft 151 KiB	1.45 T	0.66 T
stealth 198 KiB	2.75 T	1.36 T
access_rules 260 KiB	5.14 T	2.93 T
512 KiB	23.44 T	17.83 T
1 MiB	99.18 T	86.82 T

Totals include base storage and transaction weight, not premium alone.

Notes

• Consensus-breaking: publish fees change, so execution results change.
• crates/template_test_tooling and the network tables move together; FeeTable::zero_rated() leaves the flat charge at 0, so zero-rated tests are unaffected.

Testing

• cargo test -p tari_engine --lib fees — includes new coverage for the flat charge below the free allowance.
• cargo test -p tari_engine --test publish_template --test complex_fee_payment --test compute_fee_budget --test native_compute_budget — all pass.