Fix transfer reliability, chunk safety, and bounded concurrency

[daltschu22]

Dsync could return success while transfers were still running or had failed, omit files during chunking, and write/delete testfile.dsync even during cloud dry runs. This change makes the command wait for bounded parallel transfers, report failures, and preserve filename and dry-run behavior across both backends.

• Replace shell-built commands with argument lists and quoted SSH payloads; stream chunk lists over stdin so worker hosts do not need shared working files. Preserve remote path spelling instead of resolving it against controller symlinks.
• Honor --rclone-config, remove cloud write probes, avoid creating local directories for remote destinations, and fix recursive rclone basic chunking.
• Include hidden files; preserve special filenames, rsync symlinks, and empty directories. Use NUL-delimited rsync lists and raw rclone lists.
• Replace incompatible fpart -Z usage, detect nonzero failures and traversal diagnostics even when fpart exits zero, and avoid unintended recursive re-copying of fpart chunks.
• Stream basic chunk generation, skip empty jobs, enforce -n during reuse, and lock working directories.
• Handle SIGINT and SIGTERM during partitioning and transfers. Register new processes before delivering pending cancellation, finish cleanup despite further signals, and kill surviving local process-group members after a five-second grace period even if their leader exited.
• Stage chunk generation and validate reuse against source identity, mode, and checksums. Replace only verified, unchanged chunks listed in a valid manifest; preserve unrecognized files and refuse symlinked chunk directories, including during dry runs. Reject overlapping local source/destination/work paths.
• Refresh documentation and add a Python 3.10/3.14 CI matrix.

Validation: python3 -m unittest discover -s tests -v — 39 tests passed, with no skips, on Python 3.14.4, rsync 3.4.1, fpart 1.7.0, and rclone 1.60.1-DEV. Tests include real local transfers with both chunking modes and both backends, a configured rclone alias, special filenames, dry runs, unreadable directories, failure propagation, ownership checks, reuse, concurrency, locking, SIGTERM during partitioning and transfers, termination-resistant descendants, cancellation during process registration/cleanup, remote path preservation, and worker execution through a local SSH shim. git diff --check and Python compilation also passed. Actual SSH hosts and external cloud services were not exercised; large-directory throughput has not been benchmarked.

Compatibility: regenerate legacy chunks once before using --reuse; reuse requires matching backend/chunking flags and does not discover source-tree changes. Modified/unrecognized working files are preserved and require review or a new working directory. For older rclone compatibility, newline/carriage-return filenames fail explicitly before cloud transfer. Host files currently support DNS/IPv4/SSH aliases, not IPv6 literals. SSH filesystem destinations must be absolute; overlap validation covers local destinations. Remote worker cleanup depends on SSH and the remote tool's disconnect behavior. Rclone retains its default symlink/empty-directory behavior. Fpart's nonverbose stderr is checked against its normal partition statistics; unexpected diagnostics stop the run rather than risk a partial copy being reported as successful.