Keep String response bodies reusable and refuse to resend a spent streaming body

[quinnj]

Fixes #1333.

Problem

A Response built from a String body wrapped it in a cursor-based BytesBody, so the first send consumed it. Returning that same "baked" response a second time wrote a head promising Content-Length bytes and then failed: the client saw a truncated body (serve!), a 500 (streamhandler, whose adapter noticed the closed body), or an empty DATA frame (HTTP/2). A Vector{UInt8} body was stored as-is and never had the problem, which is the asymmetry the issue describes.

Reproduced with the issue's script: with serve!, /string answered once and then failed with truncated fixed-length HTTP/1 body; /bytes worked repeatedly.

Change

• Response stores String bodies as-is (_response_body_arg(::AbstractString) returns the String), exactly like Vector{UInt8} bodies. Every write path already emits AbstractString bodies zero-copy (HTTP/1 fixed-length and chunked, HTTP/2 fast path, stream adapter, display), so this also drops an allocation per response. The two positional Response(status, ::AbstractString) forms, which copied the string into a BytesBody, now go through the same rule. Requests keep BytesBody, since Request requires an AbstractBody.
• Genuinely single-use bodies are refused before the head is written: _check_response_body_unsent throws for a BytesBody/CallbackBody that was already sent or closed, or a BytesBody shorter than the declared Content-Length. write_response! and the HTTP/2 server's streaming path call it, and the request-handler server catches it and answers 500 instead of dropping the connection mid-response. Other AbstractBody implementations are left to their own semantics.
• Server guide gains a "Reusing Responses" note; CHANGELOG entry.

Tests

• http_core_tests.jl: String bodies are stored by identity (keyword, positional, compat and SubString forms); write_response! of the same response twice produces identical bytes; a spent BytesBody response throws ProtocolError (consumed) / ArgumentError (closed) before writing anything; a closed body with a declared empty payload still sends.

• http_server_http1_tests.jl: one String-body and one Vector{UInt8}-body response served twice each on both serve! and listen!(streamhandler(...)); a BytesBody response answers once and then 500.

• http2_server_tests.jl: a String-body response served twice over one HTTP/2 connection.

• http_handlers_tests.jl: the existing "streamhandler reused String response body" test pinned the old behavior (second request 500); it now asserts the second request succeeds with the same body.

• http_forms_tests.jl/http_multipart_tests.jl: the multipart helpers accept String (and any AbstractVector{UInt8}) response bodies.

Full Pkg.test on Julia 1.12.6 (macOS) passes: 29 files, Testing HTTP tests passed.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 92.00000% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.65%. Comparing base (dcf6428) to head (d736f6a).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/http_core.jl	87.50%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1364      +/-   ##
==========================================
+ Coverage   89.51%   89.65%   +0.13%     
==========================================
  Files          31       31              
  Lines       12594    12648      +54     
==========================================
+ Hits        11274    11340      +66     
+ Misses       1320     1308      -12     

☔ View full report in Codecov by Harness.
📢 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.