NarIndexer: Implement skip

[xokdvium]

Motivation

This improves the performance of parseNarListing, which is used by commands like nix nar ls when the underlying source allows cheap seeks (like StringSource or FdSource that does lseek).

For nix nar ls of a NAR for linux source tarball this cuts down the runtime almost in half (from 300ms -> 175ms).

Context

---

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[Radvendii]

change makes sense to me, and it indeed seems to speed things up a lot

poop -d 30000 '../master/result/bin/nix nar ls ../foo.nar /' '../feature/result/bin/nix nar ls ../foo.nar /'
Benchmark 1 (147 runs): ../master/result/bin/nix nar ls ../foo.nar /
  measurement          mean ± σ            min … max           outliers         delta
  wall_time           205ms ± 4.92ms     202ms …  252ms          9 ( 6%)        0%
  peak_rss           37.4MB ±  200KB    36.9MB … 38.0MB          0 ( 0%)        0%
  cpu_cycles          192M  ± 4.01M      186M  …  212M          11 ( 7%)        0%
  instructions        341M  ±  136K      340M  …  341M          18 (12%)        0%
  cache_references   57.3M  ± 1.32M     56.4M  … 67.5M           9 ( 6%)        0%
  cache_misses        399K  ± 7.53K      380K  …  426K           1 ( 1%)        0%
  branch_misses       755K  ± 6.49K      746K  …  801K           9 ( 6%)        0%
Benchmark 2 (683 runs): ../feature/result/bin/nix nar ls ../foo.nar /
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          43.9ms ± 1.78ms    40.4ms … 48.6ms          0 ( 0%)        ⚡- 78.6% ±  0.2%
  peak_rss           37.1MB ±  204KB    36.5MB … 37.7MB          2 ( 0%)          -  0.9% ±  0.1%
  cpu_cycles         98.8M  ± 1.66M     95.6M  …  108M           5 ( 1%)        ⚡- 48.6% ±  0.2%
  instructions        239M  ± 17.1K      239M  …  239M          15 ( 2%)        ⚡- 30.0% ±  0.0%
  cache_references   4.63M  ± 79.8K     4.50M  … 5.61M          32 ( 5%)        ⚡- 91.9% ±  0.2%
  cache_misses        265K  ± 6.33K      250K  …  290K           4 ( 1%)        ⚡- 33.7% ±  0.3%
  branch_misses       496K  ± 6.86K      487K  …  566K           8 ( 1%)        ⚡- 34.3% ±  0.2%

Wall time is -78.6%

[xokdvium]

@myclevorname cc. Noticed this while looking at your updated benchmarks and was wondering if we could improve anything. Turns out this content skipping wasn't working properly due to this silly mistake.