[ml service] Add Flare as a new nnfw type

[songgot]

• Implemented support for 'flare' nnfw type in ML service API
• Included test cases to validate flare functionality

[myungjoo]

Remove unrelated commits. Make it possible to be reviewed and merged independently from other topics

[songgot]

Remove unrelated commits. Make it possible to be reviewed and merged independently from other topics

Thank you for taking the time to review.
I've updated.

[hj210]

The commit purpose and changes are clearly described, and the test code appears to be well-written. Here are a few suggestions to improve maintainability and code quality:

[myungjoo-bot]

Automated review (transcribed from an AI review agent's report; please verify before acting).

Summary: The PR adds a temporary Tizen-internal #define ML_NNFW_TYPE_FLARE 23, registers "flare" in ml_nnfw_subplugin_name[], special-cases fw_name == "flare" in ml_single_open_custom(), adds a pre-switch .bin extension check in _ml_validate_model_file(), and generalizes the llama.cpp ml-service scenario test. I checked the diff against current main, merge-ability, CI status, and whether main already has an equivalent type (it does not: no FLARE / QUICKAI in main).

1. [High] Merge conflict / stale base — gh reports CONFLICTING; git merge-tree conflicts in c/src/ml-api-inference-single.c (PR ~2083-2097) and tests/capi/unittest_capi_service_extension.cc (PR ~394-541). main has since introduced _is_valid_extension() and switch ((int) *nnfw) (2a038f8), and replaced the llama.cpp test macro with LLAMACPP_TEST_MODEL / skip_llamacpp_tc (acb6d92). Since main's switch ((int) *nnfw) already removes the -Werror=switch obstacle, the FLARE check should now be a normal case ML_NNFW_TYPE_FLARE: using _is_valid_extension (model[0], { ".bin", NULL }), as @jaeyun-jung requested, instead of the pre-switch block and its @todo.
2. [High] Android CI is red for this PR only — all four build (arm64-v8a / armeabi-v7a / x86 / x86_64) jobs failed while neighbouring PRs passed. Logs have expired so I could not confirm the cause; the prime suspect is strcasecmp at c/src/ml-api-inference-single.c:1065 — use g_ascii_strcasecmp like the rest of the file. Please rebase and get CI green.
3. [Medium] Flare-specific hack instead of generic resolution — ml-api-inference-single.c:1062-1067 hard-codes "flare" before validation. _ml_get_nnfw_type_by_subplugin_name ("flare") already returns 23, so the generic fix is if (nnfw == ML_NNFW_TYPE_ANY && info->fw_name) nnfw = _ml_get_nnfw_type_by_subplugin_name (info->fw_name);, which also helps any other framework whose model extension is not auto-detectable.
4. [Medium] Temporary API placement — c/include/nnstreamer-tizen-internal.h:22: a bare #define shadows the ml_nnfw_type_e namespace and has no (Since X.0) Doxygen; the stated migration condition (after Tizen 10.0 M2) has passed and @myungjoo already asked for this to be moved into ml_nnfw_type_e in ml-api-common.h with ACR. Please do that in this PR (public API doc change included), or document who/when removes the define.
5. [Medium] Tests never run in CI and are mis-gated — scenarioConfigFlare is inside #if defined(ENABLE_LLAMACPP), and there is no Flare subplugin in CI, so the new validation/open path has zero coverage. skip_llm_tc only checks sflare_if_4bit_3b.bin but the config also needs history_lora.bin and data/tokenizer.json (partial assets fail instead of skip), and its skip message still points to the Llama-2 URL. Please add model-free unit tests that do run in CI: _ml_get_nnfw_type_by_subplugin_name ("flare") == ML_NNFW_TYPE_FLARE, its reverse, and _ml_validate_model_file with ML_NNFW_TYPE_FLARE rejecting non-.bin files with ML_ERROR_INVALID_PARAMETER; gate the scenario test on its own condition and check all three asset files.
6. [Low] Malformed test prompt — tests/test_models/data/flare_input.txt contains literal two-character \n sequences, a stray trailing ", and no trailing newline.
7. [Low] Style — ml-api-inference-single.c:2097 @todo comment is indented ~40 spaces; unittest_capi_service_extension.cc adds <iostream> / std::cout.write where the file uses g_print (use g_print ("%.*s", ...)); gchar *input_file should be const gchar *; the g_strdup ("flare_input.txt") is unnecessary; stray blank lines at ~409/474/542; the 40 s blocking g_usleep should be a bounded poll on tdata->received.

No back-door or suspicious behavior found.