Visibility of files loaded with ly:load, location of "local" guile modules

[Urs Liska]

One final question before I start working on some package code
proposal:

All the basic Scheme files are loaded in lily.scm with

  (for-each ly:load init-scheme-files)

What does this do internally? I.e. how is the visibility handled of
definitions/bindings from names in files loaded with ly:load ?

When I manually ly:load a .scm file I can see both variables created
with define and define-public. I have the impression that everything
loaded with ly:load is visible to any .ly file afterwards, is that
correct?

So my approach would be to create a file packages.scm, add it to the
list of auto-loaded files in lily.scm and define in this only what is
to become the public interface of the package mechanism. All internal
code would be loaded from there with (use-modules), or am I missing
something here?

Where would I store Scheme modules which are *not* to be loaded through
ly:load in lily.scm but through (use-modules)? 
We *do* have a number of such modules in the scm directory.

  (git grep "(scm "

indicates this for the following modules:

 * accreg
 * clip-region
 * coverage
 * display-lily
 * editor
 * framework-eps
 * framework-null
 * framework-ps
 * framework-scm
 * framework-svg
 * graphviz
 * guile-debugger
 * lily
 * ly-syntax-constructors
 * memory-trace
 * output-ps
 * output-socket
 * output-svg
 * page
 * paper-system
 * ps-to-png
 * safe-utility-defs
 * song
 * song-util
 * to-xml

So I could simply store my guile modules there too, but wouldn't it
make sense to move them to a new directory to disentangle ly:load from
use-modules files? That would seem in line with moving the .ly files
like bagpipe.ly which are not loaded automatically but only upon
request to packages.

Urs