fix(spv): compute required proof headers from actual header difficulties

Problem

The SPV maintainer computed the number of block headers for a proof assuming every header carries the relay epoch difficulty: with txProofDifficultyFactor = 1 it always assembled a single-header proof. On testnet4 (BIP94), sweep transactions are regularly mined in minimum-difficulty (DIFF1) blocks, so the proof contained only a DIFF1 header. The Bridge skips DIFF1 headers when binding to the relay difficulty, finds no decisive header, and reverts with:

execution reverted: Not at current or previous difficulty

Because the error is deterministic, the maintainer crash-looped on the same transaction every cycle (error while maintaining SPV: ... restarting maintainer), and the affected wallet's main UTXO could never be registered on the host chain — blocking all subsequent sweeps for that wallet.

Fix

getProofInfo now mirrors the Bridge's BitcoinTx.determineRequestedDifficulty / evaluateProofDifficulty logic instead of assuming a fixed per-header difficulty:

• walk headers forward from the proof start block,
• skip leading minimum-difficulty headers, but only when both relay epoch difficulties are above minimum (so test/dev relays at difficulty 1 keep working),
• bind the requested difficulty to the first decisive header matching the relay's current or previous epoch difficulty; if it matches neither, skip the transaction (the Bridge would revert),
• accumulate headers (DIFF1 headers still contribute their work) until total observed difficulty reaches requestedDifficulty × txProofDifficultyFactor,
• bound the walk by maxProofHeaders = 144 and report "wait for more confirmations" when the chain tip is reached first.

The old epoch-boundary block-count arithmetic is removed; epoch spans are handled implicitly by summing actual header difficulties.

Testing

• TestGetProofInfo rewritten for the new semantics: current/previous-epoch proofs, epoch-spanning proofs with difficulty drop/raise, leading DIFF1 headers, relay at minimum difficulty, decisive-header mismatch, DIFF1 run exceeding the header bound, and chain tip reached early. Headers are generated with real compact-bits encoding (blockHeaderWithDifficulty).
• go test ./pkg/maintainer/spv passes.
• Verified live on the Sepolia/testnet4 stack: a sweep stuck for 10 days (876510fd...62c8, mined in a DIFF1 block at height 137542) was proven on the first cycle after deploying this fix — submitDepositSweepProof tx 0xfeeefad32c7116727949483edfb1e38f8117d923521fd3467886a05226b82577, status 1 — and the maintainer completed its round cleanly.

Notes

• Requires a Bridge with DIFF1-aware determineRequestedDifficulty (deployed via the DIFF1 library upgrade); against an older Bridge the maintainer now skips such transactions instead of crash-looping.