fix(rebase): resolve build errors from rebase conflict resolution

Corrects merge artifacts introduced while rebasing the TurboQuant feature
branch onto upstream master:

- ggml.c: bump GGML_OP_COUNT static_assert 97 -> 98 (GGML_OP_TURBO_WHT added
  to the op enum; arrays already carry the entry, the count literal was left
  at master's value during the merge).
- llama.h / common.h: drop duplicate n_outputs_max field (the fork adds it,
  master already has it); keep master's "max outputs in a batch" field and
  ordering.
- llama-context.cpp: drop the matching duplicate n_outputs_max initializer.
- llama-vocab.h: drop duplicate get_suppress_tokens() declaration.
- mtmd/models.h + clip.cpp: drop duplicate clip_graph_gemma4uv struct and the
  duplicate PROJECTOR_TYPE_GEMMA4UA case (master already carries gemma4 vision).
- models/step35.cpp: restore upstream's complete file (no turbo content; the
  conflict resolution had dropped its MTP block, leaving trunk_flags undefined).

CPU-only build of libllama, llama-server, llama-cli and llama-quantize passes.

https://claude.ai/code/session_01HZZQMPA6D6KWgcuDsDarWa

Author: claude

common/common.h

@@ -431,7 +431,6 @@ struct common_params {
     int32_t n_keep                =     0; // number of tokens to keep from initial prompt
     int32_t n_chunks              =    -1; // max number of chunks to process (-1 = unlimited)
     int32_t n_parallel            =     1; // number of parallel sequences to decode
-    int32_t n_outputs_max         =     0; // max outputs supported by the context (0 = derive)
     int32_t n_sequences           =     1; // number of sequences to decode
     int32_t n_outputs_max         =     0; // max outputs in a batch (0 = n_batch)
     int32_t grp_attn_n            =     1; // group-attention factor

ggml/src/ggml.c

@@ -1122,7 +1122,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
     "GLU",
 };
 
-static_assert(GGML_OP_COUNT == 97, "GGML_OP_COUNT != 97");
+static_assert(GGML_OP_COUNT == 98, "GGML_OP_COUNT != 98");
 
 static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
     "none",
@@ -1234,7 +1234,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
     "glu(x)",
 };
 
-static_assert(GGML_OP_COUNT == 97, "GGML_OP_COUNT != 97");
+static_assert(GGML_OP_COUNT == 98, "GGML_OP_COUNT != 98");
 
 static_assert(GGML_OP_POOL_COUNT == 2, "GGML_OP_POOL_COUNT != 2");
 

include/llama.h

@@ -340,7 +340,6 @@ extern "C" {
         uint32_t n_batch;           // logical maximum batch size that can be submitted to llama_decode
         uint32_t n_ubatch;          // physical maximum batch size
         uint32_t n_seq_max;         // max number of sequences (i.e. distinct states for recurrent models)
-        uint32_t n_outputs_max;     // max outputs supported by the context (0 = derive from n_seq_max)
         uint32_t n_rs_seq;          // number of recurrent-state snapshots per seq for rollback (0 = no rollback) [EXPERIMENTAL]
         uint32_t n_outputs_max;     // max outputs in a ubatch (0 = n_batch)
         int32_t  n_threads;         // number of threads to use for generation

src/llama-context.cpp

@@ -3447,7 +3447,6 @@ llama_context_params llama_context_default_params() {
         /*.n_batch                     =*/ 2048,
         /*.n_ubatch                    =*/ 512,
         /*.n_seq_max                   =*/ 1,
-        /*.n_outputs_max               =*/ 0,
         /*.n_rs_seq                    =*/ 0,
         /*.n_outputs_max               =*/ 0,
         /*.n_threads                   =*/ GGML_DEFAULT_N_THREADS, // TODO: better default

src/llama-vocab.h

@@ -151,8 +151,6 @@ struct llama_vocab {
 
     const std::vector<llama_token> & get_suppress_tokens() const;
 
-    const std::vector<llama_token> & get_suppress_tokens() const;
-
     int max_token_len() const;
 
     int find_bpe_rank(const std::string & token_left, const std::string & token_right) const;

src/models/step35.cpp

@@ -1629,14 +1629,6 @@ struct clip_model_loader {
                         hparams.eps               = 1e-6f;
                         hparams.n_mel_bins        = 640;
                     } break;
-                case PROJECTOR_TYPE_GEMMA4UA:
-                    {
-                        // Encoder-free: raw 16 kHz waveform chunked into 640-sample frames.
-                        hparams.audio_chunk_len   = 0;
-                        hparams.audio_sample_rate = 16000;
-                        hparams.eps               = 1e-6f;
-                        hparams.n_mel_bins        = 640;
-                    } break;
                 case PROJECTOR_TYPE_GRANITE_SPEECH:
                     {
                         hparams.audio_chunk_len        = 0;

tools/mtmd/clip.cpp

@@ -24,11 +24,6 @@ struct clip_graph_gemma4uv : clip_graph {
     ggml_cgraph * build() override;
 };
 
-struct clip_graph_gemma4uv : clip_graph {
-    clip_graph_gemma4uv(clip_ctx * ctx, const clip_image_f32 & img) : clip_graph(ctx, img) {}
-    ggml_cgraph * build() override;
-};
-
 struct clip_graph_pixtral : clip_graph {
     clip_graph_pixtral(clip_ctx * ctx, const clip_image_f32 & img) : clip_graph(ctx, img) {}
     ggml_cgraph * build() override;