The terragrunt prefix regex used `:\s*` to strip the timestamp+module
prefix from each stdout line. The greedy `\s*` ate not only the single
separator space terragrunt inserts between `tf:` and the wrapped output,
but also the leading 2/4/6/… spaces of `terraform plan`'s own diff
indentation. Result: every `# resource will be created` and `+ field =
…` line in the rendered Change Result block lands at column 0, making
the diff unreadable.

Switch the trailing `:\s*` to `: ?` (optional single space). Now exactly
the separator space is consumed; terraform's indentation survives
intact. Verified end-to-end via the localfile notifier:

  before: `+ resource "x" "y" {`         (column 0)
  after:  `  + resource "x" "y" {`       (2-space indent preserved)
  after:  `      + id = (known after apply)`  (6-space indent preserved)

Added a regression test that asserts 2- and 6-space indents survive.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…d mode

Terragrunt run-all prefixes each leaf module's stdout with [module/path]
brackets; the root module's lines have no bracket. The previous parser
only matched the bracketed form, so after a leaf's section ended, root
lines were attributed to the prior leaf — e.g. a root IAM change would
show up under cluster-citadel-2g/regions/tokyo/shared-vpc in the
per-module summary.

Add a LogRootModule regex matching unbracketed terragrunt log lines and
flip currentModule back to root ("") when one appears, but only when
the previous module was set by a [bracket] log prefix. Module headers
of the form `Module <path>` (no log prefix) keep working as before —
unbracketed lines that follow them stay attributed to the declared
module.