Since we switched from autotools to Meson in commit
94592c14, we don't need to include
`config.h` early to properly enable large file support.  Meson passes
the required macros on the compiler command line instead of defining
them in `config.h`.

This means we can include `config.h` at any time, whenever we want to
check its macros, and there are no ordering constraints.

Add class Context which wraps the Android/Java Context class and add a
JNI wrapper for method Context.getCacheDir().

The new class Path only holds a string pointer without being
responsible for allocation/deallocation.  The FileSystem.hxx library
accepts Path arguments instead of AllocatedPath, to avoid forcing
callers to allocate another string object.

- DirectoryReader is extracted to separate header
- FileSystem.cxx/FileSystem.hxx/DirectoryReader.hxx moved to fs/ subdir
- Functions return true on success, instead of 0 (where applicable)
- ReadLink return result instead of out parameter
- UnlinkFile is renamed to RemoveFile
- CheckExists/CheckIsRegular/CheckIsDirectory are renamed
  to PathExists/FileExists/DirectoryExists

After we've been hit by Large File Support problems several times in
the past week (which only occur on 32 bit platforms, which I don't
have), this is yet another attempt to fix the issue.

The soure file client.c has nearly 1000 lines, time for splitting it
into smaller pieces to improve readability.

Prepare splitting client.c into several sources.

Make the client list management a separate sub-library.

The client code uses portable GLib I/O functions and doesn't need the
OS specific socket headers.

Several functions work with the wrong return type, this patch fixes
them.

When client_defer_output() aborts the connection to the client,
client_write_output() called client_write_deferred() anyway.  This
caused an assertion failure.  Fix it by checking for the "expired"
flag again after client_defer_output() returns.

I'm hunting down a bug where client->channel==NULL during I/O
operations.  These new assertions help avoid this kind of bug in the
future.

This patch fixes a memory leak: the fifo_buffer object was not freed
when the client connection was closed.

On both locations, the result of write() can be ignored safely.  In
event_pipe_emit_fast(), that can only be "EAGAIN", which means that
the pipe buffer is full - no further notification required.  In
client_init(), that would be a fatal connection error, which would be
caught by the next event.

This patch fixes gcc warnings.