Context

PR #653 added per-chunk SSE progress and surfaced it in the web Index tab. The mm init wizard's seed-index path (packages/memtomem/src/memtomem/cli/init_cmd.py:_seed_with_progress, line 1496-) consumes the same index_path_stream but ignores the new chunk_progress events. PR #653 left this out deliberately — see PR description "Out of scope" and the docstring note in the code.

Why this needs design

_seed_with_progress uses click.progressbar configured for file-unit length with this item_show_func:

click.progressbar(
    length=expected_total,                            # files
    label="  Seeding",
    item_show_func=lambda item: (item or "").rsplit("/", 1)[-1][:40],
)

Two problems:

1. The item_show_func extracts the path tail and truncates to 40 chars. Passing a free-form "file (32/250)" works for "/"-less inputs but the chunk count gets eaten by the truncate.
2. The bar's length is in files, so chunk_progress events have no natural advance unit. Just calling bar.update(0, label) works (advance=0, label refresh) but the wizard's UX tone is "minimal CLI noise" — fast-firing chunk labels may feel busy.

Options

1. Refresh sub-label without advance — bar.update(0, f"{file} {done}/{total}") and a wider item_show_func that handles both formats. Simplest. Risk: noisy on fast files.
2. Switch bar length to chunk units — length=expected_chunks instead of expected_total files; advance per chunk. Requires pre-counting chunks (currently only files are pre-counted). Cost-heavy.
3. Two-bar layout — outer file-bar, inner per-file chunk-bar shown only when chunks_total > N. Most informative but click.progressbar doesn't compose well.

Acceptance criteria

• Wizard seed run shows chunk-level progress for files producing more than progress_threshold chunks (default 32).
• Small files (under threshold) stay quiet — match the web Index tab behavior.
• KeyboardInterrupt cleanup still works (test: ^C mid-seed prints the resume hint, _close_bar runs).
• No regression in the multi-path aggregation summary line at the end.

Files likely touched

• packages/memtomem/src/memtomem/cli/init_cmd.py (_seed_with_progress)
• Possibly packages/memtomem/src/memtomem/cli/indexing.py if mm index (non-stream) is folded in at the same time — see feat(cli): convert mm index to stream + click.progressbar #656 for the related conversion.

References

• PR feat(indexing): per-chunk progress events for SSE indexing stream #653 — feature parent. PR description "Out of scope" section explains why this was deferred.