Update Proof of absence for 2 or more missing stems and serialize DepthExtensionPresent accordingly to match implementation in rust-verkle

[eclairss17]

Summary

go-verkle was sending only 1 OtherStem in the wire format regardless of how many missing stems shared the same PoA stem. Compact, but this made the proof unverifiable by rust-verkle, which expects one OtherStem entry per missing stem to know which PoA stem proves absence for which missing key.

go-verkle now sends N OtherStems (one per missing stem) in the wire format, matching the rust-verkle expectation.

Metric	Before Change	After Code Change	Delta
Internal PoaStems	1 (31 bytes)	1 (31 bytes)	0
Wire OtherStems	1 (31 bytes)	8 (248 bytes)	+217 bytes
Commitments	1 (32 bytes)	1 (32 bytes)	0
IPA proof	576 bytes	576 bytes	0
Total proof size	647 bytes	864 bytes	+33.5%

Discussion

1. The code change is a correctness fix, not a size optimization.
   Before the change, go-verkle was silently dropping OtherStems from the wire format when multiple missing stems shared the same PoA stem. In order to use rust-verkle, if it would receive a proof it cannot correctly map back to the queried keys.

2. The internal representation was already correct and deduplicated in both versions.
   proof.PoaStems is always 1 for N missing stems sharing the same PoA. This is unchanged. The optimization lives here, only one proof-of-absence path is computed cryptographically regardless of how many missing keys reference it.

3. The wire format now increases with the number of missing stems.
   The OtherStems list grows linearly: N missing stems → N wire entries × 31 bytes each. The cryptographic proof (commitments, IPA) does not grow, those stay constant.

4. The actual tradeoff is correctness vs. wire size.
   Verified cross-implementation proofs cost 31 bytes per additional missing stem that shares a PoA. For the mainnet 8-slot scenario, that is 217 extra bytes (33.5% total proof overhead) to gain rust-verkle compatibility.