feat: modifier plugins + ARA (#446)

* feat: add modifier base class

* feat: re-implement abliteration as a modifier plugin

* fix: adjust good/bad prompts hack to match tests

* feat: support dataset specifications containing multiple individual datasets

* feat: re-implement ARA as a modifier plugin

Arbitrary-Rank Ablation (ARA) (Weidmann 2026) was originally introduced by @p-e-w in #211

Co-authored-by: kabachuha <artemkhrapov2001@yandex.ru>
Co-authored-by: joninco <joninco@bullpoint.org>
Co-authored-by: Ashar <coder3101@users.noreply.github.com>

* feat: add tests for ARA

* feat: reduce default number of trials

* ci: remove broken "semantic-pull-request" workflow

* ci: add yet another alternative model hash

---------

Co-authored-by: kabachuha <artemkhrapov2001@yandex.ru>
Co-authored-by: joninco <joninco@bullpoint.org>
Co-authored-by: Ashar <coder3101@users.noreply.github.com>

tests/seed-oss/SHA256SUMS.linux

@@ -0,0 +1,6 @@
+39a23a1a28f68acd747373fa688bea630021ec4db2bc8dec2fd3962240fb0418 *chat_template.jinja
+35e3ad5aeaa78984629ceba6c22a4236b4c21d6959485900da2a5a98ea7e4062 *config.json
+2ec7a17f55287d482ca956f5acff44d9e88f6c7fd5ec06c7c33d641afa9af330 *generation_config.json
+0cdea9064dcbe6db666f9d42a283d664c133d4c67a2f6ea0fd863a54f522160c *model.safetensors
+3cf3a6d9520f195638a36f0194239d817de7288710bca55f1f5753de226748f7 *tokenizer.json
+388b47e61cb40f2fd51a89999053686ab4c45b40b43c0329d15645f5910d069e *tokenizer_config.json

tests/seed-oss/config.toml

@@ -0,0 +1,65 @@
+# This test case is for ARA (non-standard settings).
+# After any change related to it, this test should PASS.
+
+model = "tiny-random/seed-oss"
+model_commit = "6860befd78b678885f7a52bbf41d7fd0671af2db"
+
+seed = 12345
+print_debug_information = true
+
+batch_size = 2
+max_response_length = 10
+
+scorers = [
+    { plugin = "heretic.scorers.keyword_rate.KeywordRate", optimization = "minimize" },
+    { plugin = "heretic.scorers.kl_divergence.KLDivergence", optimization = "maximize" },
+]
+
+n_trials = 2
+n_startup_trials = 1
+
+export_strategy = "merge"
+checkpoint_action = "restart"
+trial_index = 0
+model_action = "save"
+save_directory = "model"
+
+[[response_prefix_test_prompts]]
+dataset = "mlabonne/harmless_alpaca"
+commit = "02c6a92cfcf11bb0c387334f8146d149d65b587f"
+split = "train[:5]"
+column = "text"
+
+[[response_prefix_test_prompts]]
+dataset = "mlabonne/harmful_behaviors"
+commit = "01cead01398926d81f7c52bdb790ee8cf77ebba7"
+split = "train[:5]"
+column = "text"
+
+[scorer.KLDivergence.prompts]
+dataset = "mlabonne/harmless_alpaca"
+commit = "02c6a92cfcf11bb0c387334f8146d149d65b587f"
+split = "test[:5]"
+column = "text"
+
+[scorer.KeywordRate.prompts]
+dataset = "mlabonne/harmful_behaviors"
+commit = "01cead01398926d81f7c52bdb790ee8cf77ebba7"
+split = "test[:5]"
+column = "text"
+
+[modifier.ARA]
+preserve_row_magnitudes = false
+lora_rank = 20
+
+[modifier.ARA.good_prompts]
+dataset = "mlabonne/harmless_alpaca"
+commit = "02c6a92cfcf11bb0c387334f8146d149d65b587f"
+split = "train[:5]"
+column = "text"
+
+[modifier.ARA.bad_prompts]
+dataset = "mlabonne/harmful_behaviors"
+commit = "01cead01398926d81f7c52bdb790ee8cf77ebba7"
+split = "train[:5]"
+column = "text"