[landsman]

Closes #68. A token opening several accounts left every app to match the stored IBAN against accounts() itself; moneta.accountByIban(token, iban) and me.accountByIban(iban) do it. Only an IBAN's letters and digits count, in either case, so one pasted from a PDF or a web page (no-break and zero-width spaces, dashes, dots) matches.

• null, not an exception, when the token does not see the IBAN: for a poller that is an answer (revoked for that account, or another holder's token).
• A blank IBAN, or one failing its ISO 13616 check digits, is an IllegalArgumentException before any request, so a caller's typo is not read as a revoked account.
• Two accounts with one IBAN (one per currency, possibly) are an IllegalStateException naming their ids, not a guess.
• Nothing remembered: one accounts request per call, so a revoked account shows on the next poll.
• An account without an IBAN never matches. No transactions(iban) shortcut: it would hide the null case in an exception.

Tests: a match as printed and with every separator; an IBAN the token does not see; blank and mistyped ones, which send no request; a shared IBAN; the accounts read again per call. .docs/usage.md has it under "More than one account", in Java and Kotlin.

[landsman]

Second pass, on the code rather than the threads. The five findings from the first round are all addressed: the lookup is on MonetaClient with the holder delegating, a blank IBAN is refused before any request, a shared IBAN throws, the per-call regex is gone, and the doc caveat now matches the filter. Tests and spotlessCheck pass on the head.

What is left:

1. .docs/usage.md now has two sections on one topic. Fold the new one into the existing ## More than one account.
2. account(token, iban) has the shape of transactions(token, accountId) but takes an IBAN. Passing an id returns null, which the docs say to read as revoked. Rename it.
3. "Spaces of any kind" still misses zero-width characters. Keeping letters and digits only fixes that and the dashes caveat together.
4. A mistyped IBAN is still indistinguishable from a revoked one. The mod-97 check is the blank guard's argument applied to the common case.
5. Two test nits.

None of this blocks the behaviour. 2 and 3 are worth settling before the method is published, since it is new and pre-1.0.