I have `134850146304` bytes of ram and Yuzu was saying that I had `125.59 GB`
of ram. But `125.59` is actually the amount of gi**bi**bytes I have. In
gi**ga**bytes I would have `134.9`.

Additionally, I changed the `1024 / 1024 / 1024` here into the `_GiB`
user-literals that I added a while ago(#6519).

https://www.wolframalpha.com/input/?i=134850146304+bytes

video_core/codecs: refactor ffmpeg searching and handling in cmake

common/cpu_detect: Remove CPU family and model

maxwell_to_vk: Add ASTC_2D_8X5_UNORM

qt_software_keyboard: Fix out of bounds array access

core/hid: Fix faulty analog triggers

common/input: Avoid numerous large copies of CallbackStatus

input/SDL: Update SDL hints

SDL_HINT_JOYSTICK_HIDAPI_SWITCH_HOME_LED is no longer needed thanks to new default in SDL 2.0.18.
SDL_HINT_JOYSTICK_HIDAPI_XBOX is reported to cause conflicts with native driver Xbox driver on Linux, and Xbox controllers don't benefit from hidapi anyways.

... this would resolve some edge-cases where multiple devices are
present and ffmpeg is unable to auto-supply the hw surfaces

input_poller/input_mapping: Silence several -Wdocumentation warnings

CallbackStatus instances aren't the cheapest things to copy around
(relative to everything else), given that they're currently 520 bytes in
size and are currently copied numerous times when callbacks are invoked.

Instead, we can pass the status by const reference to avoid all the
copying.

We currently do not make use of these fields, remove them for now.

Given these return void, these can be omitted.

param tags are supposed to specify the parameter name without any
quoting.

Silences several -Wdocumentation warnings.

Silences quite a bit of -Wdocumentation warnings, given the @param tag
is only intended to be used to identify function parameters, not what it
contains.

input_engine: Minor object churn cleanup

* this resolves the todo items in the CMakeLists.txt
* a version requirement check for ffmpeg is added to catch issues early
* for future-proof reasons, nasm/yasm is now only required when build on
  x86/AMD64 systems

We were unconditionally accessing the keyboard_buttons array, even if the bottom_osk_index was for the numberpad, leading to an out of bounds array access. Fix this by accessing the proper array for the current button when the index is for the numberpad.

tas_input: Minor cleanup

We don't need to copy the whole pair

Resolves a -Wextra-semi warning

clear() when empty is simply a no-op, so we can get rid of the check
here and let the stdlib do it for us.

Gets rid of indirect includes and includes only what the interface
needs.

While we're in the same area, we can also avoid performing std::stoi in
a loop when it only needs to be performed once.

This is only using the input facilities, so we don't need to use the
fully-fleged stringstream.

Same behavior, but without the potential for extra allocations.

This also means we can get rid of the dependency on <regex>

Parameters shouldn't have the colon by their name.

Prevents these values from potentially clashing with anything in other
headers.

We can make use of try_emplace() to insert values only if they don't
already exist.

We can use iterators to avoid looking up into maps twice in the getter
functions.

At the same time we can also avoid copying the ControllerData structs,
since they're 264 bytes in size.

These types are part of the InputCommon namespace.

Avoids creating copies of several object instances (some of which being
over 100 bytes in size).