lib/sh/sync.sh: db_snapshot(): copy changes before stopping DB to minimize the time that it's shut down

lib/sh/sync.sh: db_snapshot(): factored copy operation out into separate db_copy() function

lib/sh/sync.sh: added db_snapshot()

bugfix: lib/sh/sync.sh: don't unintentionally rsync-ignore explicitly-specified files

moved everything into /trunk/ to create the standard svn layout, for use with tools that require this (eg. git-svn). IMPORTANT: do NOT do an `svn up`. instead, re-use your working copy's existing files with `svn switch` (http://svnbook.red-bean.com/en/1.6/svn.ref.svn.c.switch.html).

bugfix: lib/sh/sync.sh: upload(): paths: don't dereference the path itself if it's a symlink; instead canonicalize just its parent dir. this allows syncing a specific file which is a symlink, rather than syncing the symlink's target.

lib/sh/sync.sh: upload(): .rsync_filter: also support machine-specific filters, for cases when different machines produce the same file (e.g. a log file) but only one machine's copy should be backed up

lib/sh/sync.sh: upload(): make put's $subpath option relative to the currdir instead, like the --include paths. note that $subpath unfortunately can't be used in subdirs at this point because it will cause rsync to ignore the .rsync_ignores and .rsync_filters in parent dirs, including the essential .rsync_ignore in the sync root dir.

lib/sh/sync.sh: upload(): documented that each --include path is relative to the currdir, not the root dir of the upload ($local_dir). this feature, although previously unintended, is actually better because the user can change to a subdir of the root dir and specify upload paths relative to the dir they are in. however, when invoking upload() from a script with --include paths specified, this means you need to use an absolute path (e.g. "$(dirname "${BASH_SOURCE⁰}")"/...; or the value that will become $local_dir, which for sync_upload() is $root_dir).

lib/sh/sync.sh: upload(): use directional .rsync_filter to supplement .rsync_ignore with all kinds of --filter rules. separate .rsync_filters are needed for the upload (swap=) and download (swap=1) directions because the sender and the receiver are reversed, causing asymmetric rules like protect/hide to change meaning.

lib/sh/sync.sh: upload(): usage: documented put's swap=1 flag, which downloads instead of uploads

lib/sh/sync.sh: upload(): use --exclude filters from per-dir .rsync_ignore. note that --exclude-from can't be used for this, because it is relative to the currdir, not the rsync root, and therefore also requires the .rsync_ignore to exist rather than using it only if it exists.

lib/sh/sync.sh: removed download(). use swap=1 upload, or swap=1 upload_caller, instead. this avoids having separate upload()/download() pairs for every caller of upload(), because you can instead just set swap=1.

bugfix: lib/sh/sync.sh: upload(): don't kw_params $swap because this unexports it, preventing put from seeing it. instead, use echo_vars to print it.

bugfix: lib/sh/sync.sh: upload(): only add `--exclude="**"` if there are --includes. this enables running upload() without paths to upload all files.

lib/sh/sync.sh: upload(): support passing -- options to put, which will not be run through the path->--include processing

bugfix: lib/sh/sync.sh: upload(): added missing `local args=()` initializer

bugfix: lib/sh/sync.sh: also need to --include parent dirs for each --include path

lib/sh/sync.sh: removed `pf upload` debug statement

added lib/sh/sync.sh with upload(), download()