Fractional indexing for kanban ordering

[hobbescodes]

Description

Replaces the integer column_index / index columns on task, column, project, and project_column with text-based fractional-indexing keys, so reordering an item is a single-row update from its destination's two neighbors instead of N sibling rewrites.

PostGraphile auto-derives the GraphQL types from the column types, so columnIndex / index fields flip from Int to String. The COLUMN_INDEX_ASC / INDEX_ASC orderBy enums keep working because lex sort on text matches base-62 ascending order.

Migration sequence (single PR, all three apply via bun db:migrate):

1. 0051_cooing_vulture (auto): add nullable text companion columns alongside the existing int cols.
2. 0052_backfill_fractional_indices (custom SQL): PL/pgSQL helper produces length-3 base-62 keys (header c), partitioned per parent (column_id, project_id, project_column_id, organization_id), ordered by existing int + created_at + id.
3. 0053_graceful_captain_midlands (custom SQL): cutover via ALTER COLUMN ... SET DATA TYPE text USING <companion>, drops the companions, adds the new (project_column_id, column_index) composite index.
4. 0054_collate_c_for_fractional_indices (custom SQL): forces COLLATE \"C\" on the four columns so Postgres ORDER BY uses byte order, matching the library's assumption ('Zz' < 'a0'). Without this, en_US.UTF-8 collation puts Zz after a0 and dragging an item to the top of a column persists a correct key but the server returns rows in the wrong order on refetch.

Insert-time key generation:

• DefaultColumnsPlugin: replaces hardcoded index: 0..4 with generateNKeysBetween(null, null, 5) so seeded columns get valid fractional keys.
• IDP webhook (handleOrganizationCreated): same treatment for default project columns.
• generateGraphqlSchema.ts: registers fractional-indexing in the EXPORTABLE module map so graphile-export can serialize generateNKeysBetween into the executable schema.

Coordinated deploy required: the GraphQL schema flips Int → String after this PR lands. The runa-app PR (omnidotdev/runa-app#TBD) consumes the new types. Deploy both services together; in-flight reorder mutations from older app instances will fail with type errors during the window between API restart and App restart.

Test Steps

• bun db:check clean before and after each migration
• bun db:migrate applies 0051 → 0052 → 0053 → 0054 in order, regenerates GraphQL schema, all four columns are text NOT NULL with COLLATE \"C\" (\\d task, \\d \"column\", \\d project, \\d project_column)
• SELECT proname FROM pg_proc WHERE proname = 'fi_seed_key' returns 0 rows after migration
• 'Zz'::text COLLATE \"C\" < 'a0'::text COLLATE \"C\" returns t
• After migrate, bun db:generate reports "No schema changes" (snapshot matches DB)
• bun run check && bun run build && bun test all pass
• src/generated/graphql/schema.graphql shows columnIndex: String! / index: String!, IntFilter becomes StringFilter, numeric aggregates removed for the four columns
• Creating a new project via the API seeds 5 default columns with valid fractional keys (a0, a1, …)