Implements the request in #5741: report the deduplicated size of a set of archives (e.g. all backups matching -a 'sh:userA-*'), following ThomasWaldmann's proposal there.

Two commits: the analyze feature, plus an unrelated one-liner fixing a stale borg info epilog.

Every figure is reported as source (uncompressed source data) and stored (compressed, as stored in the repository) size, plus the compression factor relating the two (stored / source).

Deduplicated size of a set of archives

For the considered (matching) set of archives:

• Deduplicated size of set — summed size of the union of chunks the set references (chunks shared within the set counted once).
• Exclusive size of set — summed size of chunks referenced only by the set and by no other archive, i.e. the space that deleting the whole set would free.
• Unreferenced chunks — chunks that no non-deleted archive references, i.e. what borg compact could free in the current state of the repository. Only their stored size is known: a chunk's source size is only recorded in the archives referencing it, and these have none.

Following the issue comment, chunks are flagged as referenced by the considered set (C) and/or by the rest of the archives (R); the exclusive size is sum(size(x) for x in C - R).

$ borg analyze -a 'sh:web'

Deduplicated size of the 2 considered archive(s)
===================================================================
Considered archives: 2 (of 4 in the repository)

                                  source        stored  compression
Deduplicated size of set:        8.00 MB       3.26 MB         0.41
Exclusive size of set:           2.00 MB     820.62 kB         0.41
Unreferenced chunks:                 n/a           0 B          n/a

Without an archive filter, the considered set is all archives - every referenced chunk is then trivially exclusive to it, so that line is suppressed and the report is labeled as the whole repository.

Decomposition by archive name (--by-name)

Archives sharing a name form a series, so a name usually groups all backups of one source; old-style archives that do not form a series have one name each and are grouped just as well.

Every chunk is counted in exactly one row, so the rows add up to the repository's deduplicated size. This answers "which name costs how much, and what would I get back by dropping it" for all names at once:

$ borg analyze --by-name

Repository decomposition by archive name
=================================================================================
4 archive(s) with 3 distinct name(s)

name                            archives        source        stored  compression
db                                     1       4.00 MB       1.64 MB         0.41
logs                                   1       3.00 MB       1.22 MB         0.41
web                                    2       2.00 MB     820.62 kB         0.41
(shared by 2+ names)                           6.00 MB       2.44 MB         0.41
(unreferenced)                                     n/a           0 B          n/a
---------------------------------------------------------------------------------
total (deduplicated)                   4      15.01 MB       6.11 MB         0.41

This needs only a single pass over the archives: bits 5..22 of the ephemeral user flags hold the archive name that first referenced a chunk (as index + 1, so 0 means unreferenced), bit 23 marks chunks referenced by more than one name. As the shared and unreferenced rows can only be determined by looking at every archive, --by-name always covers the whole repository and cannot be combined with archive filters.

Implementation

• Reuses the repository's own chunk index (repository.chunks) instead of building a second index — it already holds each chunk's stored size (obj_size). The membership flag bits are OR-ed in and the source size (0 in the repo index) is filled in from the per-archive references cache, preserving pack_id/obj_offset/obj_size via _replace. These in-memory mutations never persist: write_chunkindex_to_repo zeroes flags and size, and close() only serializes F_NEW entries (none here).
• Chunk membership and source sizes come from the per-archive references cache maintained by borg compact, so unchanged archives usually do not need to be opened/decrypted.
• The references-cache helpers (get_archive_references & friends, ArchiveReferences) are moved from archiver/compact_cmd.py into cache.py as free functions so analyze and compact share one implementation; compact keeps its previous behavior.

Notes

• The dedup-size report works for one or more matching archives; the existing directory hot-spot report still requires at least two (skipped with an info log otherwise).
• -a needs the sh: prefix for globs (borg2's default match is exact).
• Chunks of soft-deleted archives count as unreferenced (soft-delete keeps the archive's chunks and its reference cache; borg compact only keeps them when the repository is damaged, so borg undelete stays possible).
• Rows whose chunks carry no source size at all (e.g. only archive metadata, which is recorded with size 0) show n/a in the compression column rather than dividing by zero.

Tests

analyze_cmd_test.py covers: set dedup vs exclusive size with shared and unique files, the single-archive set, the whole-repository (no filter) report, the --by-name decomposition (asserting the rows add up to the total), the rejection of filters with --by-name, and the unreferenced-chunks report - including that borg compact then frees exactly the reported number of objects.

Rebased onto current master; analyze/compact/cache/info suites pass there (94 tests), mypy reports no new errors.