skip_pr_backfill=False shoud affect implicit default_pr_start calculation

[gjesse]

Summary

skip_pr_backfill=False should affect implicit default_pr_start resolution. If the bot is configured to backfill PR environments, then an unset default_pr_start should not silently resolve to 1 day ago just because previews need to be computed.

Current behavior

Docs say:

default_pr_start: Defaults to None meaning the start date is set to the earliest model's start or to 1 day ago if data previews need to be computed.

And:

skip_pr_backfill: Indicates if the bot should skip backfilling models in the PR environment. Default: True.

With skip_pr_backfill=False and default_pr_start=None, the bot can still choose the 1-day start when previews are needed. That means the PR env does perform a backfill, but only for a narrow window.

Expected behavior

When skip_pr_backfill=False, the implicit start should default to the earliest relevant model start / model default, unless default_pr_start is explicitly configured.

The “1 day ago if previews need to be computed” fallback should only apply when PR backfill is skipped or intentionally bounded.

Why

skip_pr_backfill=False communicates that we want PR environments to pre-populate the data needed for deploy. A 1-day implicit start can leave deployable snapshots sparsely populated, which moves historical interval work into /deploy.

Explicit default_pr_start should remain the override for teams that want bounded PR backfills.

Proposed rule

  if default_pr_start is set:
      use default_pr_start
  elif skip_pr_backfill is False:
      use earliest relevant model start / model default
  elif previews need to be computed:
      use 1 day ago
  else:
      use earliest relevant model start / model default