fix(accountsdb): don't hang when no peer serves an incremental snapshot

[0xzrf]

Fixes #1763

Problem

Bootstrapping against a cluster whose peers serve no incremental snapshots (e.g. a local solana-test-validator) hangs forever after the full snapshot downloads. In downloadSnapshotWithRetry, a peer with no incremental is skipped before dl_attempts is incremented, and the validator sets no timeout — the retry loop never exits, so the RPC server never starts.

Changes

• Count peers skipped because they have nothing downloadable in the current mode.
• If every valid peer is skipped for 5 consecutive rounds (~25s), return error.UnableToDownloadSnapshot so the existing fallback in downloadSnapshotsFromGossip boots from the full snapshot only.
• Rounds with no valid peers at all still retry as before.

Notes

Verified against solana-test-validator (Agave 4.1.2): previously the loop spun forever; now it gives up after ~25s and startup continues. An epoch-0 cluster then hits an unrelated error.InvalidInsert in EpochTracker.initFromManifest — will file separately.