Answers what schema version this package expects

[johnnyt]

What

Two remainders of Theme 7, from sp-b0g.

1. StatifierPersistence.Ecto.Migrations.expected_version/0 - the newest
   migration version this package knows, for a host whose schema is
   hand-written DDL rather than a delegated migration and which therefore
   has to check for itself that its tables are current.
2. A dated Note on ADR-0010 decision 9 answering whether an adapter that
   does not implement the optional input-log callbacks needs V05's table at
   all.

Why there is no assert_version!/1

The bead allows either the function or the documented reason it cannot
exist. It cannot: this package records no version marker in a repo's
schema - no versions table, no marker row, no version column (grep over
lib/ finds the string expected_version nowhere before this branch and no
marker of any other spelling). Ecto's own schema_migrations holds the
host's migration timestamps, which say nothing about which of V01..V05 a
hand-written schema matches. Growing such an assertion means starting to
write a marker - a new table, in a new migration, carried by every host
including the ones that delegate and already know - which is new stored
surface bought for a comparison the host can make itself. So
expected_version/0 is the whole deliverable, and its @doc says why.

The Note's answer, from the code

• An adapter that does not export the callbacks never touches the table,
  and its host does not need V05. Above the seam there is one write site
  (Runs.stepped/6 via Storage.append_input/4, which answers
  :not_supported without calling the adapter when
  Storage.input_log_supported?/1 is false) and one read site
  (Runs.inputs/2 via Storage.list_inputs/2). Below it, Storage.Ecto
  queries the inputs table only in append_input/3 and list_inputs/2 -
  their private helpers next_slot/2, insert_input/5 and input_rows/2
  are called from those two functions and nowhere else
  (lib/statifier_persistence/storage/ecto.ex:494, :499, :502, :520).
  Everywhere else the table is a name, not a query: Ecto.Config's
  @table_keys, KeyGenerator.table/0, the :uxid prefix map, and the
  generated Input schema. Such a host caps at V04 in both directions with
  the moduledoc's capped recipe rather than carrying an empty table.
• A host storing through Storage.Ecto needs V05 unconditionally:
  supports_input_log?/1 answers true without probing, so the first
  stepped event appends and a failed append fails the step. input_log_cap:
  is no opt-out either - the cap's last slot is still a written row. The
  Note names that as the trigger, and the fix if it fires (a declaration at
  init/1), without reopening the decision.

ADR-0010 is already accepted, so the Note is amend-by-addition:
git diff origin/main -- docs/adr/ is 36 added, 0 removed.

Tests

• test/statifier_persistence/ecto/migrations_test.exs - expected_version/0
  is 5. It moves with the map because @current_version is derived from
  @migrations (Map.keys |> Enum.max): a V06 entry moves the function to 6
  and the literal fails.
• test/statifier_persistence/ecto/sqlite_migrations_test.exs - the same
  assertion off Postgres, because the number is backend-independent by
  construction (it is derived from the migration map and reads no repo).
• Both sabotaged (expected_version/0 pointed at @initial_version): red,
  those two cases alone, "45 tests, 2 failures"; reverted from a copy taken
  beforehand, green after.

The pass-1 direction review caught the first shape of these cases: the
original second assertion (up(version: expected_version() + 1) raises) was
tautological, since validate_version!/2 bounds on the same attribute the
function returned. The cure commit derives @current_version from the map,
which is what makes "the newest key of the version map" true of the code and
not only of the docs, and drops the redundant assertion.

Provenance

• No file outside the bead's map was touched. migrations.ex's moduledoc
  gains one paragraph pointing at the new function - the bead allows the
  moduledoc as a home for the Note, so a pointer there is inside its scope.
• Siblings in this repo touch docs/adr/0006, docs/adr/0009 and the ADR
  README index; this branch touches none of them.

Gate

Full mix quality green on this HEAD: 482 of 482 tests, 95.7% coverage,
dialyzer and credo clean; mix quality.verify attested. Sabotage scan
reports no missing notes.

Refs: sp-b0g