fix(underwriter): bind the source-deposit hash to the caller's target_amount

[heifner]

Split out of #543, which fixes this as part of a broader UIC-prevalidation refactor. It is isolated here because it is a live defect on master with a small, independently-reviewable fix, and #543 still has open design discussion on unrelated sysio.uwrit reject semantics.

The defect

Both outposts fold the user's accepted target_amount into the SwapDeposit correlation hash:

• EVM — ReserveManagerLib.hashSwapDeposit(sender, sourceAmountDepot, args) packs args.targetAmount
• SVM — swap_correlation_hash(..., target_amount, target_tolerance_bps)

The underwriter's source-deposit verifier recomputed that preimage from uwreq.dst_amount.

Those were the same value until #550 split them. dst_amount is now the depot's AMM quote, re-priced at ingestion; sysio.uwrit.hpp states it is "never the caller's SwapRequest.target_amount". The recomputed hash therefore matches only when the caller's target happens to equal the depot's quote exactly — and a gap between the two is the expected case, which is what variance_tolerance_bps exists to bound. Any swap with a non-zero gap fails verification, the UWREQ never leaves PENDING, and no underwriter commits.

Observed

A live ETH→SOL underwritten swap (target_amount 98039214, dst_amount 97747972) produced 148 consecutive rejections against a single stable hash pair, until the flow timed out waiting for a CONFIRMED commitment:

underwriter: source-deposit verify failed for uwreq 11 — SwapDeposit hash mismatch (id=1):
  on-chain=fd8f16aad2443acf2847c6f49c41feac6f32feadc87682df25835ae328a76695
  recomputed=758ed0f49ec37030baebfdc586a4d57a36f88492c24fcb2758d8f5b4d8ffa2ff

Recomputing that preimage both ways confirms the split precisely:

preimage destination amount	keccak256
target_amount = 98039214	fd8f16aa…	the hash ReserveManager emitted
dst_amount = 97747972	758ed0f4…	what the verifier computed

The fix

Carry target_amount on the plugin's uw_request, read it from the uwreqs row, and use it in the preimage.

The packing was duplicated across the EVM and SVM verifiers, which is how one wrong field became two. Both now call a single source_deposit_hash helper (lifted verbatim from #543, so that PR's rebase is a no-op) whose input struct names the field target_amount — the depot's mutable quote is no longer reachable from the preimage by accident. The EVM leg also gains the explicit depositor-size check the SVM leg already had; previously it was only implied by a post-hoc buffer-length assertion the helper makes unnecessary.

Coverage

Two cases pin the preimage to the production vector above — the recomputed hash equals the SwapDeposit hash ReserveManager actually emitted, and the settlement quote demonstrably does not reproduce it.

Note the tests pin the helper's preimage layout and field semantics, not the two call sites' argument selection; the naming (target_amount vs dst_amount on the input struct) is what guards those.

Verification

• test_underwriter_plugin — 50 cases, *** No errors detected
• plugin_test — *** No errors detected
• nodeop links clean

No contracts/** changes, so contracts_unit_test is not implicated.

[huangminghuang]

Approved. As the PR description notes, these tests pin the helper's preimage rather than the two production call sites. The call-site regression test—using distinct target_amount and dst_amount values and asserting that the EVM/SVM verifier input selects target_amount—should be added in #543, which owns the broader verifier refactor.