docs: lead README and pubspec with chat template rendering

[leehack]

Puts chat template rendering first on pub.dev: README, Example tab, pubspec description and topics. Docs and metadata only; no behavior change.

What changed

• README: one-line pitch, dart pub add dinja, and a runnable Qwen2.5-style example with a system message, a user turn, a tool and add_generation_prompt. Then Why dinja, input marking, template analysis, a comparison with package:jinja, and scope. Removed the false "zero-dependency" claim and the ^1.1.0 pin.
• Example tab: example/chat_template_example.dart is now the README example, and example/README.md shows it inline with its output.
• pubspec: new description and topics; removed the stray # path comment. The lib/dinja.dart library doc also dropped "zero-dependency".

	Before	After
Description	A minimal, zero-dependency Jinja templating engine for Dart, ported from llama.cpp. Focused on security and efficiency.	Jinja engine for Hugging Face and GGUF LLM chat templates: messages, tools, generation prompts. Parity-tested against llama.cpp and Jinja2. All platforms, including web and Wasm. (178 chars)
Topics	jinja2, template, llm, security, chat-templates	jinja, template, llm, chat-templates, huggingface

Topics: pub.dev's doc/topics.yaml has no alias for any of them, so jinja2 is not merged into jinja. topic:jinja already lists jinja_minimal, and topic:jinja2 listed only dinja. security (100+ packages) and ai (100+) are too broad, and prompt is used by terminal-prompt packages. pub.dev's text search indexes only the description, the README and API symbols (mem_index.dart), not topics. So the description keeps Jinja2, and the README keeps prompt, template and a huggingface token (huggingface.co).

How each claim was verified

• 290 cases, 279 pass, 11 skipped: regenerating test/llama_cross_test.dart from llama.cpp 7fe450e1 tests/test-jinja.cpp gives no diff, and dart test test/llama_cross_test.dart reports +279 ~11. llama.cpp's own test-jinja at 7fe450e1 skips the same 11 as NotImplemented.
• Expected outputs match Jinja2 3.1.6: test-jinja -py with Jinja2 3.1.6 passes 290/290 with 0 failures.
• 45 real templates: test/fixtures/templates has 50 files with 45 distinct contents. Instrumenting Template.render and running test/templates showed that all 45 are rendered.
• README example output is the same in dinja, llama.cpp and Jinja2: the same template and variables go through test-jinja at 7fe450e1 (0 failures) and through Jinja2 3.1.6 in llama.cpp's transformers-style sandbox (match).
• Every README and example snippet is extracted, formatted, analyzed and run, and its stdout matches the output block that follows it exactly.
• llama-server uses common/jinja: tools/server calls common_chat_templates_apply, and common/chat.cpp runs jinja::runtime.
• Web and Wasm: dart test -p chrome and dart test -p chrome -c dart2wasm on test/src and test/llama_cross_test.dart both report +790 ~11. pana tags all six platforms and is:wasm-ready.
• Used by llamadart: llamadart 0.8.24 on pub.dev depends on dinja and imports it in chat_template_handler.dart.
• Unsupported tags and format: extends, block, include, import, raw and with each throw ParserException, and '%s'|format(1) renders %s.
• package:jinja: the wording comes from its README ("server-side template engine … template inheritance").

Gates

dart format clean · dart analyze no issues · dart test +892 ~11 · dart pub publish --dry-run 0 warnings · pana 160/160.