CSoftAE::GetDelay(), CSoftAE::GetCacheTime(), and
CSoftAE::GetCacheTotal() assume in their calculations that m_buffer
contains frames that have sink/encoder frame size. However, m_buffer
actually contains frames in the format received from CSoftAEStream.

This causes varying levels of A/V de-sync depending on other conditions.
On my Linux system I encounter this always when using using HDMI audio
and having the display refresh rate set at 24Hz.

Fix those functions to use the correct frame size.

- fix race in 2 threads consuming the ringbuffer (dvdplayeraudio calls flush and ca thread still pulls frames) by moving the flush to the ca thread instead - fixes crashing when switching tv channels (this was the real issue the 2 reverted commits before wanted to fix...)

- fix regression (stuck ringbuffer after 46min, 62min, 180min - depending on the stream setup). Revert "[AE/CA] - fix a possible overflow in the ringbuffer during readsize calculation (yes we hit this in pvr branch in internalflush - when zapping channels)"

This reverts commit e447c4bb.

Revert "[AE/CA] - ooppps - typo in last commit (note to self - commit on the same host you compiled it)"

This reverts commit 1c7791d9.

Get and use correct (removable://) file record from database Fixes DVD resume after resume details already exist

sync add-ons * https://github.com/opdenkamp/xbmc-pvr-addons/pull/178 * https://github.com/opdenkamp/xbmc-pvr-addons/pull/191 * https://github.com/opdenkamp/xbmc-pvr-addons/pull/192 * https://github.com/opdenkamp/xbmc-pvr-addons/pull/193

libcdio seems to allow read no more than about 10 audio sectors at once when CD audio device is connected via USB.
This patch makes XBMC read small number of sectors if default one fails. It uses more CPU but allows to use USB CD devices.
Tested on GNU/Linux x86 and RPi. (On Rpi OMXPlayer does not play CD, but I can rip tracks).

fixed: when doing a format conversion from float to 32 bit int, multiply with INT32_MAX - 127, since this is the maximum value that can be stored in 32 bit float and int, if INT32_MAX gets converted to float, it gets rounded to INT32_MAX + 1 which can cause wrap around distortion

I'm not sure why this was originally added.
I couldn't think of a reason why the Pi would want things done differently here,
so I tried without the ifdef. I can't see any difference in behaviour,
so I think we're better off removing it.

This reverts commit d5c4887a, reversing
changes made to 32b1a5ef.

Frodo English spelling corrections English/strings.po 

fixed: Streamdetails & resume-bookmark saving etc. for bluray folders di...