dbserver: serve the CDJ's TRACK FILTER

[vynulldev]

What & why

The deck's TRACK FILTER narrows its track lists by BPM, key, rating, colour and MyTag. With rekordbox, the deck sends the conditions and rekordbox does the filtering, so its TRACK list comes back narrowed. Against Vynull, ticking a condition unticked it straight away, the filter screen showed no settings, the MyTag screen was empty, and TRACK FILTER kept every track. link-spec only names these commands, so everything here is decoded from real rekordbox captures of a CDJ-2000NXS2 setting each condition.

Commands

• 0x3007 [ctx, on] turns the filter on or off.
• 0x3207 [ctx, prop, 0, len, blob] sets one condition. The blob is little-endian: on, kind, u16 count, u32 values.
• 0x3107 [ctx] returns all four conditions as 0x4004 [kind, 0, len, blob, 4], one big-endian FCND record each.
• 0x1015 [ctx, sort, parent, 0] lists the MyTag screen: the MyTag columns, or a column's tags (only tags some track carries, as rekordbox lists them). Ticked rows are typed 0x4b, the rest 0x4a, and a column row carries its AND/OR mode.
• 0x3307 [ctx, 0, column, on, mode] turns a column's MyTag condition on with mode 0x10 (AND) or 0x11 (OR). 0x3407 [ctx, 0, column, tag, add] ticks or unticks a tag.

Each player's settings are kept across sessions. The 0x3107 replies and the MyTag screen rows match rekordbox's byte for byte, checked against the captures.

Filtering (the TRACK list at menu location 1, while the filter is on, as rekordbox does)

• BPM: the deck sends a range in hundredths it works out itself (126 at ±1% sends 12474-12726), and tracks within it are kept.
• Key: the deck sends the chosen key's related set as rekordbox key indices ("Am related" sends 8A, 8B, 7A, 9A), and tracks in it are kept.
• Rating: [n, 0] keeps tracks rated exactly n stars.
• Colour: a bitmask with bit colour-1 (yellow, colour 4, is sent as 8; pink and blue as 65). Zero, meaning no colour chosen, keeps every track.
• MyTag: each column whose condition is on keeps tracks with all (AND) or any (OR) of its ticked tags.

Against the captured library, Vynull's counts match rekordbox's for every capture: 388 and 1,441 for two BPM ranges (1,283 for a third), 628 and 530 for two key sets, 2/5/3/0 for 5/4/2/1 stars, 3 each for yellow and red, and 1, 0, 26, 26, 27, 0 through the MyTag sequence.

Fixes along the way

• 0x3207 sends its condition blob as a trailing binary argument, which the reader didn't wait for. The blob was then read as a malformed message, the session dropped, and the deck unticked the filter. The reader now waits for it.
• Library track rows round BPM to hundredths, so a BPM like 124.74 isn't truncated just below a range's lower bound. They also carry the track's rating for the filter.
• The dbserver reads the tag store through a new MyTagSource, wired in main.go.

Tests replay the captured commands and check the replies byte for byte, the read of a real 0x3207 frame, and each condition's filtering. Conformance against rekordbox 7.2.19 is unchanged (2,764 of 4,250); link-spec has no TRACK FILTER suites.

Hardware testing

• Tested on: CDJ-2000NXS2: BPM, key, rating and colour conditions stay ticked and narrow the TRACK list (for example 1,283 tracks for BPM 122.2-137.8, as with rekordbox); turning a condition off restores the full list. MyTag: the screen lists the columns and their used tags, ticking Trap with Genre on (AND) keeps 1 track, adding Deep House keeps 0, switching to OR keeps 27 and unticking Trap keeps 26, as with rekordbox; a second column (one tag, AND) keeps 3. No session drops or unhandled requests. Rebased on the sort menu (dbserver: serve the deck's SORT menu and sort track lists like rekordbox #100) and rechecked with the synthetic test library (vynull-tools testlib): the related-key, rating and MyTag conditions keep exactly the expected tracks, the MyTag screen lists only used tags (an empty column lists none), and the filter turns on and off repeatedly without a session drop.

Checklist

• go build ./..., go vet ./..., and go test ./... pass
• gofmt -l . is clean
• New source files carry an SPDX header (GPL-3.0-or-later)
• Tested on real hardware (deck + firmware noted above), or this change doesn't affect deck behaviour
• I agree my contribution is licensed under the project's GPLv3