Add --import-json to compare against previously saved benchmark runs

[ChrisJr404]

Closes #607.

This adds an --import-json <file> flag that loads benchmarks from a JSON file previously written by --export-json and folds them into the current run. The imported entries are not re-executed: they appear in the relative speed comparison and in every export format alongside any commands given on the command line.

hyperfine 'old-binary args' --export-json baseline.json
hyperfine --import-json baseline.json 'new-binary args'

--import-json can be specified more than once to combine several files. It can also be used on its own (without any commands) to convert a saved JSON run into another export format, which is the use case @dkaszews mentioned on the issue.

Implementation notes

• BenchmarkResult now derives Deserialize in addition to Serialize. The fields that --export-json skips serialize as default so the existing on-disk schema parses without loss.
• A new src/import/ module reads the JSON and reconstructs BenchmarkResult values. command_with_unused_parameters is not part of the on-disk schema, so it falls back to command when missing.
• The scheduler grew a small add_imported_results entry point. Imported results are pushed onto its results vector before any live benchmark runs, so they participate in print_relative_speed_comparison and in all configured exporters.
• Live benchmarks now carry a separate display_number so the Benchmark N: headers continue counting from the imported tail rather than restarting at 1. The original number field still indexes the per-command option arrays (command_output_policies, prepare, conclude), keeping that bookkeeping unchanged.
• --command is no longer required when --import-json is supplied, so a saved file can be replayed (or re-exported) without inventing a dummy command.

Tests

• Unit tests under src/import/mod.rs: a JSON round-trip (re-emit a BenchmarkResult via serde and reload it), tolerance for missing optional fields, and a friendlier error for a missing path.
• Integration tests in tests/integration_tests.rs covering: import + a live command, import-only, multiple --import-json flags in the same invocation, and the missing-file error message.

Output looks like:

Benchmark 1: sleep 1.0 (imported)
Benchmark 2: sleep 2.0 (imported)
Benchmark 3: sleep 1.5
  Time (mean ± σ):      1.500 s ±  0.000 s    [User: 0.000 s, System: 0.000 s]
  Range (min … max):    1.500 s …  1.500 s    10 runs

Summary
  sleep 1.0 ran
    1.50 ± 0.00 times faster than sleep 1.5
    2.00 ± 0.00 times faster than sleep 2.0