Add deepseek_dsml tool parser for DeepSeek-V4's native DSML format

[snagnever]

Adds a tool parser for DeepSeek-V4 (Flash/Pro)'s native DSML tool-call format, so tool calling works for these models through the OpenAI-compatible server.

Why

DeepSeek-V4 emits tool calls in its own DSML format — not Hermes <tool_call> JSON or any of the existing parsers' formats — so there is currently no tool_parser_type that parses its output, and tool calls come back as plain content.

Format

<｜DSML｜tool_calls>
<｜DSML｜invoke name="get_weather">
<｜DSML｜parameter name="city" string="true">Paris</｜DSML｜parameter>
<｜DSML｜parameter name="days" string="false">3</｜DSML｜parameter>
</｜DSML｜invoke>
<｜DSML｜invoke name="get_weather">...</｜DSML｜invoke>
</｜DSML｜tool_calls>

• Multiple <｜DSML｜invoke> per block → native parallel calls.
• string="true" → literal string value; string="false" → JSON value (number/bool/array/object).

Change

• mlx_lm/tool_parsers/deepseek_dsml.py — new parser (modeled on minimax_m2).
• _infer_tool_parser — one entry so the official DSML chat template (deepseek-ai/DeepSeek-V4-Flash#16) auto-selects it; no manual tool_parser_type needed.
• Tests — single call, parallel calls, mixed string/JSON params.

Marker note

The start/end markers use the <｜DSML｜tool_calls prefix (dropping the trailing >). mlx-lm matches markers by exact token-id sequence, and on this tokenizer the > merges with the following byte (...calls>\n → one token), so the full marker never matches — the same class of issue as #1335. The ｜DSML｜ core is a special token, so the prefix is a stable anchor; the parser tolerates the leftover > before the first <｜DSML｜invoke>.

Verification

On mlx-community/DeepSeek-V4-Flash-2bit-DQ (no tool template → tools dropped → 0 tool calls as a baseline), adding the official DSML template + this parser: 39/40 (98%) on a jdhodges-style tool suite and 8/8 on the parallel multi-tool cases — competitive with full-size locals. python -m unittest tests.test_tool_parsing passes (existing parsers unaffected; this is purely additive).

Pairs with the official template PR #16. Independent of #1335/#1336 (different module), though it uses the same prefix-marker insight.

[nh13]

Thanks for this, @snagnever — I hit the same DSML tool-calling gap on DeepSeek-V4-Flash and independently landed on the same parser shape (parallel invokes, string="true|false", the no-trailing-> marker anchor). Two things from running it against the 4/5/6-bit MLX quants that may be worth folding in:

1. Truncated invoke close tag — those quants systematically emit </｜DSML｜inv> instead of </｜DSML｜invoke>, so the strict close in _INVOKE misses the call and it returns as content. A lenient close (</｜DSML｜inv[^>]*>) recovers it.
2. Encode side — DeepSeek ships no Jinja template (only encoding_dsv4.py), so there's currently nothing to pair the parser with under chat_template_type. I ported encode_messages to a small chat_templates/deepseek_dsml.py reusing the deepseek_v32 helpers, byte-exact for single-turn prompts.

Both are on a branch stacked on this PR (#1760, draft) if useful — glad to fold them into yours instead if you'd rather. Also worth noting: #1501 (merged) makes the trailing-> marker workaround unnecessary on current main.