feat: batch validation with auto-managed, resumable sessions

[kikkomep]

This PR adds batch validation to the CLI: validate --batch validates every RO-Crate found under a directory and aggregates the results, with progress reporting, a per-crate summary table and machine-readable outputs. Each run is recorded as an auto-managed session that is saved incrementally (one save per validated crate), so an interrupted run — Ctrl+C, a crash, a killed job — can be resumed from where it stopped, either automatically by re-running the same command or explicitly via the new sessions subcommand.

What's included

CLI — validate --batch

• --batch/-b validates every crate under RO-CRATE-URI (subdirectory crates, zipped crates and detached *-ro-crate-metadata.json files); --batch-pattern filters the crate names.
• Live progress bar on stderr (passed/failed/remaining), per-crate result lines, and a final summary table (size, status, checks, issues, duration per crate); -v adds the details of every failed crate.
• Exit code reflects the aggregated outcome (0 only if every crate passed).
• Output formats: text (console or file), json, csv (one row per crate).

CLI — sessions

• sessions list (alias ls): stored sessions with status, progress and target (--status, --json).
• sessions show <ID>: re-render the recorded output of a session without re-validating anything; --stats for the statistics, -o report.md -f md to write them to a file.
• sessions resume <ID>: continue an interrupted session (only the pending crates are processed).
• sessions clear (alias rm) and sessions path to manage the history.
• Interactive mode offers the stored sessions as a menu when no ID is given.

Sessions & resumability

• Sessions are stored under the user cache directory and keyed deterministically by scan target, profile
  selection and severity: re-running the same command resumes the same session automatically; --no-resume starts over. Changing profiles or severity intentionally starts a new session.
• The session file records per-crate outcomes (status, checks counters, serialized issues, duration, size), so summaries and reports can be rebuilt entirely from what was saved.

Memory behaviour

Batch runs keep memory usage flat regardless of the corpus size: per-crate outcomes are persisted to the session file instead of being retained in memory (retaining every live ValidationResult pinned ~4 MB/crate and got large batches OOM-killed). Live results can still be retained explicitly for debugging or scripting purposes via batch_validate(..., keep_results=True), exposed as BatchValidationResult.live_results; the recorded outcome of every crate is always available in .crates.

Usage

# Validate every crate under a directory (auto-resumes if interrupted)
rocrate-validator validate --batch path/to/crates/ -p ro-crate-1.1

# Machine-readable report + statistics
rocrate-validator validate --batch path/to/crates/ -f csv -o report.csv
rocrate-validator validate --batch path/to/crates/ --stats

# Browse and resume past runs
rocrate-validator sessions list
rocrate-validator sessions resume <ID>