ReadProjectionProgressAsync reports the persisted heartbeat and agent status (#5172)

[jeremydmiller]

Closes #5172.

The bug

Both ReadProjectionProgressAsync overloads hardcoded AgentStatus and LastHeartbeat to null and never selected the columns. The XML docs justified it:

ProjectionProgressRow.AgentStatus and ProjectionProgressRow.LastHeartbeat are always null: Marten models the columns but no daemon path writes them (#519).

That was true when it was written. Since #537 the daemon does write them — ExtendedProgressionWriter → WriteExtendedProgressionAsync populates heartbeat and agent_status on every flush, and AllProjectionProgress reads them back correctly through ProjectionProgressStatement + ShardStateSelector. Only these read paths were never updated.

The sting: #435 added the targeted per-cell read specifically so a monitor polling one visible cell wouldn't have to pull every projection × tenant row on each tick. A caller taking that supported path to ask "is this shard's agent alive" got NULL, with no way to tell a placeholder from a fact — while the row-scan path they were told to avoid returned the real value.

The fix

Both queries select agent_status, heartbeat when EnableExtendedProgressionTracking is on, and hydrate them into ProjectionProgressRow. With extended tracking off the columns don't exist on mt_event_progression, so the narrow column list is kept and both fields stay null — that's a fact ("this store doesn't track it"), not a placeholder.

In the version-collapsing overload the telemetry is captured from the row that wins the newest-version/highest-sequence contest, not from whichever candidate happened to be read first.

Kept minimal per the issue: ProjectionProgressRow has no fields for pause_reason, running_on_node or the failure_* columns, and monitors wanting those still have AllProjectionProgress.

Not in this PR

ProjectionProgressRow's own doc comment in JasperFx (src/JasperFx.Events/ProjectionProgressRow.cs:21-25) carries the same stale claim ("Neither Marten nor Polecat writes it today"). That's a JasperFx change and is left as a hand-off.

Tests

DaemonTests/Bugs/Bug_5172_read_progress_extended_columns.cs — 5 cases:

• the projection-name overload reads back the persisted status + heartbeat
• the exact-ShardName overload does too, on a tenant-bearing identity
• the winning version supplies the telemetry (an older version's stale Stopped/heartbeat must not leak through)
• an undecorated row still reports nulls
• with extended tracking off, both overloads still work and report nulls (no undefined-column failure)

All 5 pass, along with the existing 8 read_projection_progress tests (net10.0).

🤖 Generated with Claude Code

https://claude.ai/code/session_017CTtw2kVRSZKp1p5RTxgAy