#35561 net: move some CNodeState fields to Peer

• Reduce lock contention on the main engine lock during peer download and block negotiation
• Decouple peer state tracking from internal node validation locks

Moves `m_requested_hb_cmpctblocks`, `m_provides_cmpctblocks`, and `fPreferredDownload` (renamed `m_preferred_download`) from `CNodeState` to `Peer` as atomic variables. Also makes `m_num_preferred_download_peers` an atomic integer, allowing several call sites in `net_processing.cpp` to drop `cs_main` locking.

Problem: Protecting peer download and compact block negotiation flags with `cs_main` introduces unnecessary lock contention, thread safety annotations, and split state across `CNodeState` and `Peer`.

P3 · cleanup

• P3 because it incrementally reduces main lock contention without resolving an active bottleneck or bug
• Provides small leverage toward decoupling network processing from the main lock

Reduces cs_main locking scope in net_processing by moving three peer state booleans and an integer counter to atomics. While it serves the ongoing goal of decoupling net_processing from cs_main, it addresses no active bug or major performance bottleneck.

Membership: Modifies peer state structures and lock usage inside src/net_processing.cpp.

Factors: security/stability 0, bug 0, performance 1, user value 0, leverage 1

• Needs rebase due to merge conflicts with master

The PR has merge conflicts against master and is marked with the 'Needs rebase' label.

Author status: active; last push addressed review feedback and rebased, but currently silent for 22 days while conflicting with master

Resolved concerns:

• dergoegge noted that reasoning about `m_num_preferred_download_peers` without a shared lock was harder, but agreed the atomics are correct; mzumsande noted precedent with `m_wtxid_relay_peers`.
• mzumsande suggested adding a negative lock annotation `!m_peer_mutex` to `NewPoWValidBlock`, which the author added.

• Strong support for reducing main lock scope for peer state tracking (davidgumberg, mzumsande)
• Confirmed the atomic implementation is safe and matches existing patterns (dergoegge, mzumsande)
• Code approval on the latest head commit (w0xlt)

Strong support for dropping cs_main and migrating state to Peer; all reviewer questions resolved

Reviewers agreed with the direction to reduce cs_main locking scope and approved the atomic implementation across iterations without objections.

• davidgumberg Concept ACKed: 'Less cs_main is good, locks are scary, and can hurt performance, and are unnecessary for all of these fields.'
• dergoegge reviewed and ACKed an earlier revision, confirming the changes look fine and safe.
• mzumsande Concept ACKed, pointing out consistency with existing atomic patterns in net_processing.
• w0xlt ACKed the latest head commit.

Review verdicts (DrahtBot): 1 (+1)

• ACK: w0xlt
• Stale ACK: dergoegge
• Concept ACK: davidgumberg, mzumsande

4 lines under test/bench/ci.

• src/net_processing.cpp +39/-39
• src/node/txdownloadman_impl.cpp +1/-1
• test/functional/test_framework/test_framework.py +1/-1
• test/functional/wallet_listsinceblock.py +1/-1

PR 35561 moves compact block relay flags and preferred download status from CNodeState to Peer as atomic variables, and converts m_num_preferred_download_peers into an atomic integer. This removes the need for cs_main locking across several peer processing paths in net_processing.cpp. The change is internal cleanup that incrementally reduces cs_main lock contention and simplifies peer state tracking. The PR has strong review support with an ACK on the latest push, but is currently stale pending a rebase.