chore: release v0.4.15

[MagicalTux]

🤖 New release

• fstool: 0.4.14 -> 0.4.15 (✓ API compatible changes)

Changelog

0.4.15 - 2026-06-10

Added

• (create) two-phase analytic builder — FsSizePlan via FilesystemFactory, AFFS reference
• (create) writer-determined exact sizing for all block filesystems
• (create) exact content-fit sizing infra + FAT32 (FsSizePlan)

Fixed

• (hfs+) fsck-clean create at any size (alt-VH alignment + bitmap padding)

Fixed

• (hfs+) create -t hfsplus now produces fsck.hfsplus-clean volumes at
  any size, not only block-aligned ones. Two pre-existing writer bugs:
  (1) the image could be a non-multiple of the 4 KiB allocation block (the
  2× + 64 MiB auto-size, or an arbitrary --size), leaving a trailing
  partial block past the alternate volume header where fsck reads a
  misplaced header — HfsPlus now reports image_len, so the create/repack
  paths truncate the output to a whole number of allocation blocks; (2) when
  the bitmap's last byte was partial, its padding bits were written as 1,
  but TN1150 requires bits beyond total_blocks to read as 0 — they're now
  cleared on disk (the in-memory allocator still keeps them set). Verified
  fsck.hfsplus-clean across previously-failing block counts and 500/5000-file
  trees.

Added

• (create) content-fit sizing, exact per filesystem. fstool create <fs> <source-dir> without --size now computes the minimal image that
  holds the content, instead of a 2× + 64 MiB over-provision (or, for FAT32,
  requiring --size). A new FsSizePlan trait (mirroring ext's BuildPlan)
  lets each filesystem accumulate the exact on-disk allocation from the single
  analysis walk and return the smallest image its writer accepts, rounded only
  to that filesystem's native unit. FAT32 is the first wired up: it
  searches the authoritative Fat32::geometry, so e.g. 250 MiB of content
  produces a ~254 MiB image (≈1 % overhead, all but one cluster used) and a
  source-backed create -t fat32 no longer needs --size.
• (create) the writer now determines its own size for every block
  filesystem (hfs+, hfs, affs, xfs, ntfs, f2fs). Rather than a parallel size
  model that could drift, create <fs> <dir> (no --size) does a dry-run:
  it formats and populates the real writer against a sparse, write-discarding
  SizingDevice — assigning inodes/CNIDs, encoding names, building B-trees and
  directory blocks exactly as for a real build, with file data written as
  zeros so the probe is metadata-only — and binary-searches the smallest size
  the writer's own allocator accepts. Result: tight, fsck-clean images
  (hfs +0 %, affs +1 %, hfs+ +3 %, xfs +13 %; ntfs and f2fs report their
  writers' genuine minimums) versus the former 2× + 64 MiB. Compressing /
  archive backends (squashfs, iso, grf, …) keep their grow-then-truncate path.
  The probe tolerates writers that panic on degenerate small sizes (caught
  and treated as "doesn't fit"), e.g. a pre-existing f2fs format panic.

---

This PR was generated with release-plz.