test_pageserver_gc_compaction_smoke: could not find data for key

[arpad-m]

Originally reported by @alexanderlaw in: #10391 (comment):

I've reproduced these scary errors with test_pageserver_gc_compaction_smoke too, by running 8 test instances in parallel, even on 4c4cb80 (dated 2024-12-09): compute.log:

PG:2025-01-22 16:19:17.514 GMT [976688] ERROR:  [NEON_SMGR] [shard 0] could not read block 162 in rel 1663/5/16384.0 from page server at lsn 0/09BB7660
PG:2025-01-22 16:19:17.514 GMT [976688] DETAIL:  page server returned error: Read error

pageserver.log:

2025-01-22T16:18:51.405445Z  INFO wal_connection_manager{tenant_id=584a20da29656b33e10cbb8ce14da0e1 shard_id=0000 timeline_id=5d722464ed806f311b1748d2964f9a0a}:connection{node_id=1}:tokio_epoll_uring_ext::thread_local_system{thread_local=12 attempt_no=0}: successfully launched system
...
2025-01-22T16:19:17.514097Z ERROR page_service_conn_main{peer_addr=[::1]:55676}:process_query{tenant_id=584a20da29656b33e10cbb8ce14da0e1 timeline_id=5d722464ed806f311b1748d2964f9a0a}:handle_pagerequests:request:handle_get_page_at_lsn_request_batched{req_lsn=FFFFFFFF/FFFFFFFF}: error reading relation or page version: Read error: whole vectored get request failed because one or more of the requested keys were missing: could not find data for key 000000067F000000050000400000000000A2 (shard ShardNumber(0)) at LSN 0/9BB7661, request LSN 0/9BB7660, ancestor 0/0
...
2025-01-22T16:19:18.602413Z  INFO wal_connection_manager{tenant_id=584a20da29656b33e10cbb8ce14da0e1 shard_id=0000 timeline_id=5d722464ed806f311b1748d2964f9a0a}:connection{node_id=1}: walreceiver connection handling ended: connection closed

[skyzh]

The issue seems to only happen when gc-compaction decides to produce an image that covers the full range 0000..FFFF. In theory this should not happen b/c no historic layer should cover that full range.

[skyzh]

I suppose there are some places in the code that incorrectly assumes 000..FFF are L0 layers (even if I added the logic to allow 000...FFF image layers)