Add specification for double delta filter encoding. (#5137)

[SC-49776](https://app.shortcut.com/tiledb-inc/story/49776)

This PR adds the encoding of the double delta filter to the filter
specification. The specification was based on the filter class'
[documentation](https://github.com/TileDB-Inc/TileDB/blob/6a9918a6caf33967dc3201016ecb51eb62a45d3c/tiledb/sm/compressors/dd_compressor.h#L61-L102)
and
[implementation](https://github.com/TileDB-Inc/TileDB/blob/6a9918a6caf33967dc3201016ecb51eb62a45d3c/tiledb/sm/compressors/dd_compressor.cc#L209-L257).

---
TYPE: FORMAT
DESC: Added specification for the encoding of the double delta filter.

Author: teo-tsirpanis

format_spec/filters/double_delta.md

@@ -0,0 +1,32 @@
+---
+title: Double Delta Filter
+---
+
+## Double Delta Filter
+
+The double delta filter compresses integer type data by first computing the delta between consecutive elements, then the delta between the deltas, and bit-packing the result.
+
+### Filter Enum Value
+
+The filter enum value for the double delta filter is `6` (`TILEDB_FILTER_DOUBLE_DELTA` enum).
+
+### Input and Output Layout
+
+The input data layout will be an array of integer numbers (each known as `in_{n}`, with `n` starting from 0). Their type (henceforth known as `input_t`) is inferred from the output type of the previous filter, or the tile's datatype if this is the first filter in the pipeline, but can be overriden by the [_Reinterpret datatype_ field](../filter_pipeline.md#delta-compressor-options) in the filter options.
+
+The output data layout consists of the following fields:
+
+|Field|Type|Description|
+|:---|:---|:---|
+|`bitsize`|`uint8_t`|Minimum number of bits required to represent any `dd_n` value.|
+|`n`|`uint64_t`|Number of values in the input data.|
+|`in_0`|`input_t`|First input value.|
+|`in_1`|`input_t`|Second input value.|
+|`sign_2`|`bit`|Sign of `(in_2 - in_1) - (in_1 - in_0)`.|
+|`dd_2`|`bit[bitsize]`|Absolute value of `(in_2 - in_1) - (in_1 - in_0)`.|
+|…|…|…|
+|`sign_n`|`bit`|Sign of `(in_n - in_{n - 1}) - (in_{n - 1} - in_{n - 2})`.|
+|`dd_n`|`bit[bitsize]`|Absolute value of `(in_n - in_{n - 1}) - (in_{n - 1} - in_{n - 2})`.|
+|`pad`|`bit[((n - 2) * (bitsize + 1)) % 64]`|Padding to the next 64-bit boundary.|
+
+If `bitsize` was computed as equal to `sizeof(input_t) * 8 - 1` (i.e. double delta compression would not have yielded any size savings), double delta compression is not applied and the input data will be added to the output stream unchanged, after `bitsize` and `n`, which are always written.