bip38, tool, bench, random: check dogecoin_random_bytes everywhere

[xanimo]

Eleven call sites discarded the return value of dogecoin_random_bytes — six in BIP38, the rest in cli/tool.c, src/random.c, src/bench.c, and the OP-TEE and OpenEnclave hosts.

This matters most at bip38.c's seedb: seedb derives factorb, and the private key is passfactor * factorb mod N. With the return ignored, an RNG failure produced a key from whatever was already in the buffer.

#382 and #405 make that survivable by zeroing the buffer on failure — but survivable isn't correct. A caller that can't distinguish a key from an empty buffer will still use it.

How each site propagates

Site	Approach
bip38_encrypt_ec_with_passphrase, dogecoin_bip38_encrypt_from_intermediate, hd_gen_master	already returned a status — check and return false
bip38_ownerentropy_generate, make_valid_privkey	were static void — now dogecoin_bool, callers propagate
OP-TEE / OpenEnclave hosts	sit directly in main() — report and exit non-zero

Two couldn't change signature:

dogecoin_bip38_generate_lot_sequence is LIBDOGECOIN_API void. Rather than break the signature it emits lot 0 — which every consumer already rejects (lot == 0 is checked at the top of both encrypt paths). So an RNG failure surfaces through validation that already exists.

random_seed is reached through a void function pointer in fast_random_context. It now leaves requires_seed set and doesn't key the RNG, so a later call retries rather than the context coming up keyed with an all-zero seed and reporting itself ready.

The test took two attempts to prove anything

The first version installed a failing RNG mapper and asserted the encrypt paths refused. It passed even with the seedb check reverted — because the owner-entropy check fails first and shadows it. It was testing the aggregate, not the site.

The isolating assertion uses dogecoin_bip38_encrypt_from_intermediate with the BIP38 spec intermediate code, so the owner entropy is already fixed and the RNG is reached for seedb alone.

With that in place, reverting the seedb check doesn't merely fail an assertion:

seedb reverted:  rc=139   (SIGSEGV)

The code carries on with an unwritten seedb and crashes downstream. Restored: 82/82.

---

Worth noting there are now no unchecked dogecoin_random_bytes calls left in the tree.

This will show red on cppcheck until #403 lands — that job fails on 0.1.5-dev itself, not on this branch.