Add consensus block size limit

[mtg0001]

Summary

• Add MAX_BLOCK_SIZE = 1000000 as the consensus block-size ceiling.
• Make CBlock::CheckBlock() validate serialized block size against that ceiling instead of the generic 32 MB serializer limit.
• Add CBlock::CheckSizeLimits() so the rule is tested directly.
• Add a consensus-limits self-test and include it in both Windows and Linux test targets.

Why

MAX_SIZE is the serializer safety limit, not a block consensus rule. This tree comes from Bitcoin 0.1.0, before Bitcoin added the later 1 MB block cap, so Bitflash had no real block-size ceiling beyond 32 MB.

The current chain audit found the largest real block at 34 KB across 5,709 blocks. A 1 MB cap leaves wide room for existing usage while preventing oversized blocks from becoming valid consensus history.

Consensus warning

This is a consensus rule change. Do not merge only from unit tests. Before merging, run a full validation from genesis with the proposed rule, for example a fresh sync or a full -checkblocks=0 validation against the current chain, and confirm no existing block is rejected.

I did not run that against the production data directory in this workspace because doing so would touch the local node/wallet state. This PR is prepared for audit.

Testing

• git diff --check
• make -f makefile.mingw tests from MSYS2 UCRT64 in src/

[Bitflash-sh]

Merged as #111 — your commit, rebased over #109 with the self-test conflict resolved (both PRs add a self-test and a line to each makefile's tests target, and git folded the two new functions into one).

The resync is the part I want on the record, since this is consensus. A node built from the rebased branch synced genesis → 5706 on the live network: zero size-limit rejections, zero CheckBlock errors. And the cap was checked against the chain before trusting the number — 5709 blocks read out of blk0001.dat give a largest block of 34,290 bytes against a median of 216, so 1 MB leaves 29× of headroom and rejects nothing that exists.

One thing I did not take: vtx.size() <= MAX_BLOCK_SIZE compares a transaction count against a byte limit. It is inherited from 0.1.0 and inert — a transaction serializes to more than sixty bytes, so a million of them cannot weigh under a megabyte. I left the line and wrote down why beside it: removing it would be tidier and would change nothing, which is exactly the argument that talks people into rewriting consensus code.

Closing this one in favour of #111. Thanks — this and #109 both landed today.