perf: reduce EventSourceStream parser allocations

[trivikr]

This relates to...

Fixes: #2630

Rationale

EventSourceStream was doing avoidable work on long-lived SSE streams by concatenating the buffered input for every chunk, allocating strings for every parsed field name, and replacing the internal event object after each dispatch. That creates unnecessary allocation churn in the hot path, especially when servers deliver heavily fragmented chunks.

Changes

• Replaced the single-buffer Buffer.concat([buffer, chunk]) approach with chunk-list buffering plus cursors.
• Kept line parsing lazy so buffers are only concatenated when a logical line spans multiple chunks.
• Matched data, event, id, and retry field names from bytes instead of decoding field strings first.
• Validated retry and id directly from bytes before decoding their values.
• Reset the parser event state in place instead of allocating a fresh event object on every event boundary.
• Added focused EventSourceStream benchmarks covering realistic payloads, fragmented chunking, and BOM-prefixed input.

Features

N/A

Bug Fixes

Reduced allocation pressure in the EventSourceStream parser hot path for fragmented SSE streams.

Breaking Changes and Deprecations

• No public API changes.
• Internally, processEvent(event) now receives a reused parser event object instead of a freshly allocated object each time.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready

[trivikr]

Benchmarks

Before

$ node benchmarks/eventsource/eventsource-stream.mjs
clk: ~3.18 GHz
cpu: Apple M1 Pro
runtime: node 24.14.0 (arm64-darwin)

benchmark                   avg (min … max) p75 / p99    (min … top 1%)
------------------------------------------- -------------------------------
• EventSourceStream parsing (250 events)
------------------------------------------- -------------------------------
single chunk                 523.13 µs/iter 517.21 µs █                    
                    (493.29 µs … 853.88 µs) 731.96 µs █                    
                    (167.25 kb …   1.18 mb) 659.19 kb █▆▂▁▁▂▃▃▂▂▁▁▁▁▁▁▁▁▁▁▁

256-byte chunks              544.53 µs/iter 538.29 µs █                    
                      (519.42 µs … 1.71 ms) 749.08 µs █▆                   
                    (110.61 kb …   2.11 mb) 711.91 kb ██▆▂▂▂▂▂▂▁▁▂▁▂▁▁▁▁▁▁▁

64-byte chunks               597.89 µs/iter 600.50 µs  █                   
                    (580.54 µs … 844.83 µs) 726.08 µs ▆█                   
                    (148.83 kb …   1.41 mb) 868.50 kb ██▇▇▄▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁

8-byte chunks                  1.14 ms/iter   1.14 ms  █ ▃                 
                        (1.11 ms … 1.30 ms)   1.27 ms  █▇█▅                
                    (659.22 kb …   2.29 mb)   2.29 mb ▆████▆▃▂▂▂▂▁▁▁▁▂▂▃▂▁▂

1-byte chunks                  5.49 ms/iter   5.56 ms ▃█▆▇                 
                        (5.20 ms … 6.90 ms)   6.63 ms █████▃▂              
                    ( 13.86 mb …  13.86 mb)  13.86 mb ███████▆▂█▁▁▃▁▂▁▂▁▁▂▂

• EventSourceStream parsing with BOM
------------------------------------------- -------------------------------
BOM + 8-byte chunks            1.16 ms/iter   1.17 ms  █                   
                        (1.13 ms … 1.39 ms)   1.34 ms  █ ▄                 
                    (  2.29 mb …   2.30 mb)   2.29 mb ▆███▅▅▂▂▂▁▁▁▁▁▁▁▁▁▁▂▁

BOM + 1-byte chunks            5.42 ms/iter   5.49 ms  █▂                  
                        (5.32 ms … 5.74 ms)   5.70 ms  ██                  
                    ( 13.86 mb …  13.87 mb)  13.86 mb ▃██▆▂▆▄▂▂██▂▃▅▁▁▂▁▂▂▂

After

$ node benchmarks/eventsource/eventsource-stream.mjs
clk: ~3.14 GHz
cpu: Apple M1 Pro
runtime: node 24.14.0 (arm64-darwin)

benchmark                   avg (min … max) p75 / p99    (min … top 1%)
------------------------------------------- -------------------------------
• EventSourceStream parsing (250 events)
------------------------------------------- -------------------------------
single chunk                 356.42 µs/iter 333.46 µs █▇                   
                      (320.71 µs … 6.80 ms) 661.83 µs ██                   
                    ( 10.12 kb …   1.12 mb) 210.89 kb ██▃▄▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁

256-byte chunks              364.56 µs/iter 363.38 µs   █                  
                    (352.96 µs … 517.63 µs) 443.25 µs ███                  
                    ( 62.13 kb …   1.30 mb) 301.20 kb ███▆▂▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁

64-byte chunks               465.56 µs/iter 463.33 µs  █▅                  
                    (445.42 µs … 650.08 µs) 588.88 µs ▅██                  
                    ( 14.05 kb …   1.09 mb) 552.85 kb ███▆▄▂▂▂▂▂▂▁▂▁▁▁▁▁▁▁▁

8-byte chunks                  1.01 ms/iter   1.00 ms    █                 
                      (976.13 µs … 1.22 ms)   1.13 ms  ▆▄█                 
                    (  1.08 mb …   2.11 mb)   1.83 mb ▆████▄▃▂▂▂▁▁▁▁▁▁▂▂▂▂▁

1-byte chunks                  5.10 ms/iter   5.10 ms  █                   
                        (4.84 ms … 6.46 ms)   6.43 ms  █ ▃                 
                    ( 11.93 mb …  11.93 mb)  11.93 mb ▄███▄▃▁▁▂▁▁▁▂▁▁▁▂▁▂▁▁

• EventSourceStream parsing with BOM
------------------------------------------- -------------------------------
BOM + 8-byte chunks            1.02 ms/iter   1.02 ms   █                  
                      (993.00 µs … 1.33 ms)   1.19 ms   █                  
                    (  1.84 mb …   1.88 mb)   1.84 mb ▄▃██▃▃▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁

BOM + 1-byte chunks            5.13 ms/iter   5.20 ms     ▆█               
                        (4.96 ms … 5.37 ms)   5.36 ms     ██▄ ▄   ▃        
                    ( 11.93 mb …  11.93 mb)  11.93 mb ▅██▅█████▇█▆█▆▇▇▅▅▅▅▂

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.05%. Comparing base (0860142) to head (21710c4).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5032      +/-   ##
==========================================
+ Coverage   93.02%   93.05%   +0.02%     
==========================================
  Files         110      110              
  Lines       35793    35888      +95     
==========================================
+ Hits        33298    33396      +98     
+ Misses       2495     2492       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.