feat: Show detailed information for one run in the terminal (#251)

[anna495]

Summary

Add areno show <run_id> — display model, dataset, algorithm, resolved key settings, current stage, latest metrics, and last error for a single training run. Reads from existing local artifacts with no new dependencies.

Usage

areno show <run_id>                  # human-readable output
areno show <run_id> --format json    # structured JSON
areno show <run_id> --format table   # compact key-value table

Run ID can be a full ID, partial ID prefix (if unique), or PID.

CLI Output

Run: train sft Qwen/Qwen3.5-0.8B  (id=841950e0f0ed)
Kind:     train
Status:   exited
Stage:    train_start
Step:     1027
Created:  2026-07-30T06:07:24+00:00
Updated:  2026-07-30T06:22:21+00:00
Metrics:  /tmp/areno/tfevent_sft

Key settings:
  algo                   sft
  ckpt                   Qwen/Qwen3.5-0.8B
  dataset_path           yahma/alpaca-cleaned
  world_size             2
  batch_size             2
  mini_bs                2
  optimizer_lr           1e-06
  lr_decay_style         cosine
  weight_decay           0.01
  ...

Latest metrics (29):
  rollout/accuracy               0.0  @step 1026
  rollout/prompt_len_mean        25.0  @step 1026
  rollout/response_len_mean      54.5  @step 1026
  train/loss                     1.4218  @step 1026
  train/grad_norm                45.300  @step 1026
  ...

Timing (80 steps recorded):
  Avg total / step:    1.16s
  Avg train / step:    1.16s

Recent logs (last 1 lines):
  registered AReno command: areno train --ckpt Qwen/Qwen3.5-0.8B ...

What changed

File	Change
areno/cli/show.py	New: areno show CLI with human/table/json output
areno/cli/main.py	+1 line: register show command
tests/test_cli_show_cpu.py	New: 20 CPU tests
docs/show-run.md	New: user guide with usage, examples, limitations

Edge cases handled

Scenario	Behavior
Run ID not found	Clear error, exit code != 0
Ambiguous partial ID	Lists matching IDs, exit code != 0
Partial ID prefix	Matches if unique
Missing metrics	Section omitted, no error
Missing config	Section omitted, no error
Partially written artifacts	Still produces output with available fields
Active/running run	Shows "running" status and current stage, no exit code
Secrets in config	Auto-redacted (key/token/password/credential)
Full samples	Never printed

Design decisions

• Dashboard API first, local fallback: Tries GET /api/jobs/{id} first for full data, falls back to reading local artifact files when server is unavailable.
• Sections config normalization: Dashboard API returns config with a sections key for structured display. _normalize_dashboard_config() extracts flat key-value pairs for CLI consumption.
• Secret redaction: Config keys matching key/secret/token/password/credential patterns are automatically hidden from output.
• Partial ID matching: Users can type a prefix of the run ID instead of the full 12-character hex.

Tests (20 CPU)

Category	Count	Coverage
Human format	7	Basic output, key settings, metrics, error section, timing, secrets redaction
JSON format	1	Valid JSON with expected fields
Table format	1	Key fields present in table output
Invalid input	6	Nonexistent ID, ambiguous partial, partial match, empty metrics/config, partial artifacts
Active runs	2	Running status, stage display, no exit code
Resolve job	3	Exact ID match, PID match, no match

Related Issue

Closes #251