Add get_key() builtin for table value-to-key lookup

[borisbat]

Summary

• Add get_key(tab, value_ref) builtin that computes a table key from a value reference obtained during values() iteration — O(1) pointer arithmetic on the parallel key/value arrays
• Robust validation: bounds check, alignment check, tombstone/deleted-slot check with proper error messages via Context/LineInfo
• Install utils/perf_lint/ in CMake install targets
• Fix perf_lint.rst exit code documentation (was "1 if warnings", actually "1 if compile errors")
• Add benchmarks/core/table/ comparing get_key vs keys()+values() iteration
• Add benchmarks/core/array/test02.das — push vs push_clone benchmark for int, struct, and large struct (10K elements, pre-reserved)
• Reorganize benchmarks: rename array_lock/ → array/, table_key_value/ → table/
• Add roadmap/optimizations.md with optimization notes on each_kv iterator and push vs push_clone
• Replace 10 handmade doc stubs with proper descriptions (back, each, emplace, empty, find_index_if, get, insert_default, key_exists, lock, subarray)

Test plan

• bin/Release/daslang.exe dastest/dastest.das -- --test tests/language/table_get_key.das — 12 tests (int/float, string/int, const table, tombstones, empty, single entry)
• Error handling: passing a non-table value pointer correctly throws via try/recover
• Benchmark compiles: dastest/dastest.das -- --test benchmarks/core/table/test01.das
• Benchmark compiles: dastest/dastest.das -- --test benchmarks/core/array/test02.das
• das2rst.das regenerated, no stubs remain, no Uncategorized sections
• CI