fix(assert): Render header values readably; document multi-value

[korya]

Problem

A failed header assertion printed Go's slice syntax at the user, and nothing anywhere said that a header can hold more than one value.

$ http-assert --assert-header-eq 'X-Custom: wrong' http://.../ok
- header[X-Custom]: expected "wrong", got ["abc123"]
                                          ^^^^^^^^^^ Go slice syntax, for a header with one value

That bracket was also the only place the tool admitted the model is a list. --assert-header and --assert-header-eq hold when any value matches, so this passes against a response that sets three cookies:

$ http-assert --assert-header-eq 'Set-Cookie: session=abc' https://example.com/login
[+] PASSED

The flag said "Assert header equals the provided value" and the README said "exact value". Both read as a claim about the only value, and neither --help nor the README mentioned multi-valued headers at all.

Solution

Render values the way a person writes them, and write the matching rule down in the three places a user would look.

Before / after, for the same two responses:

- header[X-Custom]: expected "wrong", got ["abc123"]          →  got "abc123"
- header[X-Dup]: expected "wrong", got ["first" "second"]     →  got "first", "second"

The semantics are unchanged — the ticket is right that any-value is the sensible default. What changes is that --help, the flag descriptions and the README now say so, deliberately placed next to the existing paragraph about repeating the flag, which is a different thing and was the only nearby text a reader could mistake for this one.

The reported flag was one of three

--assert-header and --assert-header-missing formatted the same []string the same way, so the bug had three manifestations, not one:

Assertion	Before
--assert-header-eq	expected "nope", got ["a=1" "b=2"]
--assert-header	expected to match "^nope$", got ["a=1" "b=2"]
--assert-header-missing	expected to be missing, got ["a=1" "b=2"]

Fixing only the reported one would have left two assertions rendering disagreements with it.

Verification

New e2e tests cover the any-value rule for both matching flags, the strictness of --assert-header-missing, both renderings, and a guard that no header failure contains got [ again. They were mutation-checked: with headerValues reverted to %q on the slice, five of the seven subtests fail.

Manually reproduced the ticket's exact commands against a server with X-Dup: first / X-Dup: second and three Set-Cookie headers, before and after.

Other Changes

• No behaviour change. The only observable difference is the text of a failure message; exit codes, matching and everything else are untouched.
• Eight unit expectations that pinned the bracketed form were updated, plus the --assert-header-missing regex that matched got \[.*\].
• No visual change — CLI only.

Closes #97

🤖 Generated with Claude Code