fix(report): keep blueprint-status' count columns when the first row is NOT_DEPLOYED

[neilmartin83]

Fixes #356.

The bug

A table's columns are the keys of its first row, so a NOT_DEPLOYED blueprint sorting first took SUCCEEDED, FAILED and PENDING off the whole report — while -o json showed the counts on the rows below it, which is what made it read as a rendering fault rather than a missing column. The reporter saw it across three environments; the only predictor was whether the first blueprint was DEPLOYED.

The fix

Per command rather than in the formatter, matching the config list precedent — unioning keys centrally would change every table in the CLI, which CLAUDE.md rules out.

Every row now carries every key, with nil meaning "not applicable to this row", and blueprintStatusRowsForFormat resolves that per format:

• a structured format (json, json-multi, yaml, ndjson) drops the key, keeping the shape those formats have always emitted
• every column-rendering format substitutes an em dash, so the key survives as a column whatever the first row happens to be

An em dash rather than 0, because 0 succeeded devices and no deployment to have succeeded on are different facts. -o xml and -o raw count as column-rendering: neither has a case in the formatter's switch, so both render a table and both need the column.

Two adjacent silences, now reported

GetBlueprint and GetBlueprintReport failures were swallowed with if err == nil, so a failed detail fetch left the row claiming no scope and no steps — and if it failed on the first blueprint it took SCOPE and STEPS off the report too, by the same mechanism. Both now warn to stderr and leave the cell not-applicable.

Tests

TestReportBlueprintStatusRendersEveryColumnFromTheWire drives the command over the reported wire shape through an httptest server, so it fails if the command stops carrying a key on a row that has no count — the resolver tests take the canonical rows as given and cannot see that. Verified failing against the previous row build, with exactly the reported output.

Wire verification

Live EU environment, 23 blueprints, first two rows NOT_DEPLOYED — the shape the issue needs. Before and after run against the same tenant minutes apart.

Before (main):

 NAME                                         SCOPE   STATE          STEPS
                                              0       ○ NOT_DEPLOYED     1
 TEST                                         1       DEPLOYED       1

After:

 NAME                                         FAILED   PENDING   SCOPE   STATE          STEPS   SUCCEEDED
                                              —        —         0       ○ NOT_DEPLOYED     1       —
 TEST                                         0        0         1       DEPLOYED       1       1
 Claude Coder and Desktop Governance Policy   3        2         1       DEPLOYED       1       0
 OS updates                                   0        6         8       DEPLOYED       1       0

All six formats exercised, exit 0 and empty stderr on each (so neither new warning fired — every fetch succeeded):

format	result
table	7 columns, — placeholders
csv	header + 23 rows, all 7 fields
plain	23 rows, 7 fields each
json / yaml / ndjson	0 em dashes; NOT_DEPLOYED rows carry 4 keys, DEPLOYED rows 7

go build, full go test ./... and make lint (0 issues) all clean.

🤖 Generated with Claude Code