fix: do not create change or split outputs below Bitcoin Core dust limit

[limpbrains]

Problem

Change created when remainder > 148 * feeRate. At 1 sat/vB that is 149 sats. Bitcoin Core dust limit is 546 (p2pkh) / 540 (p2sh) / 294 (p2wpkh) / 330 (p2wsh, p2tr). Tx with such change rejected by nodes as dust. Same for split (send max) outputs. Known upstream, never fixed: bitcoinjs#86, bitcoinjs#16.

Fix

• New utils.isDust(value, output, feeRate): dust if value <= 148 * feeRate (not worth spending, strict as before) or value < Core dust limit for output type (inclusive, same as Core IsDust). Core limit = same math as GetDustThreshold(), dustrelayfee 3 sat/vB. No script = p2pkh. Used by finalize (change) and split.
• New optional 4th arg options for all algos + finalize:
  • changeScript: { length } - real size of change output + its dust limit. Omitted = p2pkh, old behaviour.
  • txExtraBytes: number - tx bytes lib does not know about, e.g. 1 for segwit marker & flag.
• Invalid options (wrong types, unknown keys like { length: 34 }) = no solution {}. Silently ignoring them burned whole change as fee. utils.transactionBytes returns NaN for them. changeScript.length over 10000 (Core MAX_SCRIPT_SIZE) rejected too.
• blackjack overpay tolerance stays 148 * feeRate. Tying it to dust limit made it burn up to 546 sats when solution with change was cheaper (test added).
• index.d.ts, README updated.

Backwards compatible: 3-arg calls behave the same outside sub-dust window. Below-dust change goes to fee instead.

Check

• npm test: 258 pass, standard clean, coverage 100%. No old fixture changed.
• Differential fuzz old vs new, 300k random cases, half steered to dust boundary: 258354 identical. 41646 differ, all the same way: change < 546 dropped, same inputs, fee up by exactly that change. 0 other diffs. 0 cases where old found solution and new did not. Dust change 0.
• With options (248k solved): change never below limit of its type, fee never below feeRate * size.
• split with exact script lengths: 668 cases old solved and new did not, all 668 were real dust (0 relayable).

Notes

• Witness output detected by script length (22 / 34), lib only gets length. Unknown length = non-witness = higher limit = safe side. Caller must pass exact scriptPubKey length, padded length (e.g. +3) gets the non-witness limit.
• Known leftover: dropping sub-dust change can cost more than adding one more input would. Old code made unrelayable tx there. Follow-up for accumulative.

[GladosBlueWallet]

Test subject, your dust patch does what the commit message promises—how refreshingly rare. utils.isDust and relayDustThreshold now enforce Bitcoin Core dust limits (plus the existing spend-cost check) in finalize and split; sub-dust scraps get incinerated into fees instead of spawning relay-rejected outputs. changeScript and txExtraBytes propagate through every algorithm; checkOptions returns {} on bad input. blackjack keeps its separate 148 * feeRate overpay tolerance—two experiments, one facility. Three-argument callers stay compatible outside the sub-dust window. README and index.d.ts document the new behavior. Tests hit real selection/finalize paths with concrete values, not mocks. Diff trace: no regressions. Proceed. For now.