v4.10.1

A patch release fixing two defects in the 4.10.0 MySQL CLI work, both found by
running a real end-to-end MySQL installation.

Bug Fixes

• @byline/cli — byline init could not wire a host application's own
  vite.config.ts, leaving a freshly installed app unable to boot. The installer
  wrote its canonical Vite config only when the file was absent, byte-identical
  to canonical, or hashed to a previous Byline config. A stock TanStack Start
  application matches none of those, so every first-time install fell through to
  a warning — and init treats anything short of a blocking error as a soft
  warning, then went on to report "installation complete" over an application
  that failed at request time with
  ERR_UNKNOWN_FILE_EXTENSION ".css". The upgrade path had always worked, which
  is why this went unnoticed; the first-install path never had.

  Byline's required settings are now merged into the host's existing config
  rather than the file being left alone. The merge is deliberately conservative:
  each insertion must land in a key the host has not already claimed, a key
  holding a host-authored value is reported for you to merge by hand rather than
  overwritten, and a config with no inline defineConfig({ ... }) object still
  blocks — with the required settings printed, which the blocking path previously
  omitted. The original file is preserved under a timestamped
  vite.config.ts.bak-<stamp> name, which also retires a trap in the previous
  fixed vite.config.bak name: a second run found that file already present and
  silently degraded to a manual instruction.

  The settings themselves are extracted from Byline's canonical template at
  runtime rather than duplicated into the merge logic, so the template remains
  the single source of truth, and a contract test fails loudly if a piece is
  renamed or removed there.

• @byline/cli — scaffolded seed-admin, seed-docs, and import-docs
  scripts hung after completing their work instead of exiting. The configured
  adapter's connection pool holds a live timer, so the process stayed alive long
  after the seed had committed. pgAdapter's two-second idle timeout masked this
  on PostgreSQL — the scripts appeared to exit on their own — while mysqlAdapter
  leaves mysql2's defaults in place, whose sixty-second idle sweep is not
  unref'd, so a MySQL installation looked frozen. The reference application was
  corrected for this previously; the templates the CLI ships were not. All three
  entry points now exit explicitly, and a failed run exits non-zero rather than
  surfacing an unhandled rejection.

Chores

• @byline/cli — the super-admin seed now logs only that it is running and
  what the outcome was, matching the shape the example documents seed already
  used, instead of enumerating which of role, user, and role-assignment it
  created.

All other @byline/* packages bumped to 4.10.1 in lockstep with no
behavioural changes this cycle.