feat: drop frozen filed in BytesBuf

[Vaiz]

BytesBuf::frozen is used only in ensure_frozen() and in debug asserts, there is not much value in caching it and keeping it up to date

[Copilot]

Pull request overview

Removes the cached BytesBuf::frozen field and computes frozen length on demand to simplify invariants and state updates.

Changes:

• Dropped the frozen: usize struct field and its update sites.
• Added a frozen() helper to compute frozen length from len and the (logically) first span builder.
• Updated debug output/assertions and adjusted the BytesBuf size test accordingly.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[sandersaares]

there is not much value in caching it and keeping it up to date

Can you quantify this with benchmark before/after numbers?

[Vaiz]

there is not much value in caching it and keeping it up to date

Can you quantify this with benchmark before/after numbers?

hmm, I got a steady 5% drop on consume_many_spans bench with 50'000 samples

 cargo bench -p bytesbuf --quiet --bench buf --all-features  -- consume_* --baseline consume_50000_og
BytesBuf/consume_one_span
                        time:   [95.115 ns 95.233 ns 95.355 ns]
                        change: [−2.7936% −2.6326% −2.4536%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1665 outliers among 50000 measurements (3.33%)
  822 (1.64%) high mild
  843 (1.69%) high severe
BytesBuf/consume_max_inline_spans
                        time:   [86.757 ns 86.819 ns 86.883 ns]
                        change: [+0.0717% +0.1750% +0.2800%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 4613 outliers among 50000 measurements (9.23%)
  16 (0.03%) low mild
  2083 (4.17%) high mild
  2514 (5.03%) high severe
BytesBuf/consume_many_spans
                        time:   [196.09 ns 196.38 ns 196.70 ns]
                        change: [+4.5611% +4.7347% +4.9335%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 2202 outliers among 50000 measurements (4.40%)
  2 (0.00%) low mild
  506 (1.01%) high mild
  1694 (3.39%) high severe