server: add encoder-decoder model support (T5, BART, MADLAD)

[Turee]

Add support for encoder-decoder models in llama-server, matching the behavior of llama-cli. This enables translation models like MADLAD and other T5-based models to work with the server.

Changes:

• Add has_encoder flag to detect encoder-decoder models at load time
• Implement llama_encode() call for encoder-decoder prompt processing
• Use decoder_start_token to initialize decoder after encoding
• Clear decoder KV cache before each new request (no prefix caching)
• Disable incompatible features for encoder-decoder models:
  • Context shift (encoder outputs are fixed)
  • Speculative decoding (not supported)
  • Warn about prompt caching differences
• Add edge case handling for empty text tokens

The encoder processes the full prompt, then the decoder generates output using cross-attention to the encoder's hidden states.

Make sure to read the contributing guidelines before submitting a PR

Cursor /w Opus 4.5 was heavily utilized to develop this change.

[Turee]

Ran cpu only tests:

100% tests passed, 0 tests failed out of 38

Label Time Summary:
main = 268.73 sec*proc (38 tests)

[Turee]

Performance seems unaffected

(add-enc-dec-model-support)> ./build/bin/llama-bench -m ~/Downloads/gemma-7b.Q2_K.gguf
ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
ggml_metal_library_init: using embedded metal library
ggml_metal_library_init: loaded in 0.007 sec
ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
ggml_metal_device_init: GPU name:   Apple M3 Max
ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
ggml_metal_device_init: GPU family: MTLGPUFamilyMetal3  (5001)
ggml_metal_device_init: simdgroup reduction   = true
ggml_metal_device_init: simdgroup matrix mul. = true
ggml_metal_device_init: has unified memory    = true
ggml_metal_device_init: has bfloat            = true
ggml_metal_device_init: has tensor            = false
ggml_metal_device_init: use residency sets    = true
ggml_metal_device_init: use shared buffers    = true
ggml_metal_device_init: recommendedMaxWorkingSetSize  = 28991.03 MB
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| gemma 7B Q2_K - Medium         |   3.24 GiB |     8.54 B | Metal,BLAS |      10 |           pp512 |        503.37 ± 0.50 |
| gemma 7B Q2_K - Medium         |   3.24 GiB |     8.54 B | Metal,BLAS |      10 |           tg128 |         45.38 ± 0.06 |

build: 80305701 (7362)

 ((a81a5695))> ./build/bin/llama-bench -m ~/Downloads/gemma-7b.Q2_K.gguf
ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices
ggml_metal_library_init: using embedded metal library
ggml_metal_library_init: loaded in 0.010 sec
ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s)
ggml_metal_device_init: GPU name:   Apple M3 Max
ggml_metal_device_init: GPU family: MTLGPUFamilyApple9  (1009)
ggml_metal_device_init: GPU family: MTLGPUFamilyCommon3 (3003)
ggml_metal_device_init: GPU family: MTLGPUFamilyMetal3  (5001)
ggml_metal_device_init: simdgroup reduction   = true
ggml_metal_device_init: simdgroup matrix mul. = true
ggml_metal_device_init: has unified memory    = true
ggml_metal_device_init: has bfloat            = true
ggml_metal_device_init: has tensor            = false
ggml_metal_device_init: use residency sets    = true
ggml_metal_device_init: use shared buffers    = true
ggml_metal_device_init: recommendedMaxWorkingSetSize  = 28991.03 MB
| model                          |       size |     params | backend    | threads |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --------------: | -------------------: |
| gemma 7B Q2_K - Medium         |   3.24 GiB |     8.54 B | Metal,BLAS |      10 |           pp512 |        503.40 ± 0.31 |
| gemma 7B Q2_K - Medium         |   3.24 GiB |     8.54 B | Metal,BLAS |      10 |           tg128 |         45.03 ± 0.26 |

build: a81a5695 (7361)

[Turee]

Perplexity before:
Final estimate: PPL = 3672.3405 +/- 94.20817
After:
Final estimate: PPL = 3672.3405 +/- 94.20817