Also contains const correctness changes.

The FILTER_PRED function is templated and has two separate instantations
for 10 and 12 bit separately. (They're switched between using a
runtime check on entry to the function.)

This allows testing cases where this function internally switches
between two implementations for those two, without adding a bpc
argument to the init function (forcing testing and benchmarking every
16 bpc ipred function twice).

Before:                Cortex A53     A72     A73
cfl_ac_420_w4_8bpc_neon:    131.8    75.6    70.8
cfl_ac_420_w8_8bpc_neon:    199.4   106.4   117.8
cfl_ac_420_w16_8bpc_neon:   370.6   194.6   213.3
cfl_ac_422_w4_8bpc_neon:     98.4    61.4    56.6
cfl_ac_422_w8_8bpc_neon:    237.7   134.2   141.0
cfl_ac_422_w16_8bpc_neon:   456.5   256.2   279.5
After:
cfl_ac_420_w4_8bpc_neon:    121.1    76.3    67.2
cfl_ac_420_w8_8bpc_neon:    188.7   106.6   115.3
cfl_ac_420_w16_8bpc_neon:   331.7   177.4   199.8
cfl_ac_422_w4_8bpc_neon:     88.7    57.3    51.6
cfl_ac_422_w8_8bpc_neon:    208.2   121.2   130.7
cfl_ac_422_w16_8bpc_neon:   393.8   226.3   239.3

In these cases, we only need the value of the first element.

This is around one cycle faster.

cfl_ac_444_w4_8bpc_c: 499.1
cfl_ac_444_w4_8bpc_ssse3: 24.3
cfl_ac_444_w4_8bpc_avx2: 28.9
cfl_ac_444_w8_8bpc_c: 1240.2
cfl_ac_444_w8_8bpc_ssse3: 47.4
cfl_ac_444_w8_8bpc_avx2: 34.9
cfl_ac_444_w16_8bpc_c: 1785.7
cfl_ac_444_w16_8bpc_ssse3: 86.7
cfl_ac_444_w16_8bpc_avx2: 54.6
cfl_ac_444_w32_8bpc_c: 4343.5
cfl_ac_444_w32_8bpc_ssse3: 236.5
cfl_ac_444_w32_8bpc_avx2: 113.6

h_pad and w_pad can only be even if ss_ver=0 or ss_hor=0, respectively.
This means certain special cases don't need to be implemented in SIMD
while still guaranteeing correct decoding, and thus we don't want to
test for these special cases in the checkasm test either.

gen_grain_uv_ar0_8bpc_420_c: 72275.4
gen_grain_uv_ar0_8bpc_420_ssse3: 7274.8
gen_grain_uv_ar0_8bpc_422_c: 111742.9
gen_grain_uv_ar0_8bpc_422_ssse3: 13724.8
gen_grain_uv_ar0_8bpc_444_c: 205688.5
gen_grain_uv_ar0_8bpc_444_ssse3: 26218.3
gen_grain_uv_ar1_8bpc_420_c: 100682.5
gen_grain_uv_ar1_8bpc_420_ssse3: 20168.4
gen_grain_uv_ar1_8bpc_422_c: 167931.4
gen_grain_uv_ar1_8bpc_422_ssse3: 39524.7
gen_grain_uv_ar1_8bpc_444_c: 323812.2
gen_grain_uv_ar1_8bpc_444_ssse3: 77930.3
gen_grain_uv_ar2_8bpc_420_c: 159545.7
gen_grain_uv_ar2_8bpc_420_ssse3: 25849.7
gen_grain_uv_ar2_8bpc_422_c: 295959.9
gen_grain_uv_ar2_8bpc_422_ssse3: 49286.6
gen_grain_uv_ar2_8bpc_444_c: 571862.2
gen_grain_uv_ar2_8bpc_444_ssse3: 98814.2
gen_grain_uv_ar3_8bpc_420_c: 243445.9
gen_grain_uv_ar3_8bpc_420_ssse3: 28806.2
gen_grain_uv_ar3_8bpc_422_c: 458189.9
gen_grain_uv_ar3_8bpc_422_ssse3: 56629.9
gen_grain_uv_ar3_8bpc_444_c: 883627.3
gen_grain_uv_ar3_8bpc_444_ssse3: 114761.2

Also contains slight fixes to generate_grain_uv.ar0 to not pack before
adding the current grain value. Fixes overflows in e.g. seed=1115072968.

gen_grain_uv_ar0_8bpc_420_c: 72275.4
gen_grain_uv_ar0_8bpc_420_avx2: 7253.4
gen_grain_uv_ar0_8bpc_422_c: 111742.9
gen_grain_uv_ar0_8bpc_422_avx2: 13704.1
gen_grain_uv_ar0_8bpc_444_c: 205688.5
gen_grain_uv_ar0_8bpc_444_avx2: 25007.5
gen_grain_uv_ar1_8bpc_420_c: 100682.5
gen_grain_uv_ar1_8bpc_420_avx2: 18434.4
gen_grain_uv_ar1_8bpc_422_c: 167931.4
gen_grain_uv_ar1_8bpc_422_avx2: 37817.9
gen_grain_uv_ar1_8bpc_444_c: 323812.2
gen_grain_uv_ar1_8bpc_444_avx2: 74049.6
gen_grain_uv_ar2_8bpc_420_c: 159545.7
gen_grain_uv_ar2_8bpc_420_avx2: 23994.0
gen_grain_uv_ar2_8bpc_422_c: 295959.9
gen_grain_uv_ar2_8bpc_422_avx2: 48103.5
gen_grain_uv_ar2_8bpc_444_c: 571862.2
gen_grain_uv_ar2_8bpc_444_avx2: 93044.6
gen_grain_uv_ar3_8bpc_420_c: 243445.9
gen_grain_uv_ar3_8bpc_420_avx2: 27698.3
gen_grain_uv_ar3_8bpc_422_c: 458189.9
gen_grain_uv_ar3_8bpc_422_avx2: 54183.1
gen_grain_uv_ar3_8bpc_444_c: 883627.3
gen_grain_uv_ar3_8bpc_444_avx2: 103296.7

Also contains slight fixes to generate_grain_uv.ar0 to not pack before
adding the current grain value. Fixes overflows in e.g. seed=1115072968.

For armv7 & aarch64

This fixes the following warning:
../src/decode.c(666): warning C4244: '=': conversion from 'uint16_t' to 'uint8_t', possible loss of data

This allows combining two MV comparisons at once for compound refs.

Approximately 12.1% faster while using 27.7% less memory on first 1000
frames of Chimera 8bit/1080p on a Haswell system, single-threaded.

Process input data as soon as it's fed to dav1d_send_data() instead of storing
a single packet and expecting the user to call dav1d_get_picture() in order to
trigger the decoding process before they are allowed to feed more data.

Availability is sketchy if _FILE_OFFSET_BITS is defined to 64 which
meson does by default.
See https://android.googlesource.com/platform/bionic/+/master/docs/32-bit-abi.md#32_bit-and

The callback function may be NULL, not Dav1dSettings field.

All other branches are built but not deployed.

The description was being added only to the last field of each line by Doxygen.

The argument for --input was aligned with the argument for
--output. None of the other arguments were aligned.
For consistency either align all or none.
This commit removes the alignment.

avx512 was merged with avx512icl.
See 7b208fa8

This currently does not check the vulkan/placebo codepath since needed
packages are not yet in Debian unstable.