fix(context): accept USER.md-style names for show/pull/push

[eddietejeda]

Normalizes the context name argument for show, pull, and push: trim, take the final path segment, then strip a trailing .md (ASCII case-insensitive) so autocomplete paths like USER.md or ./DATAMODEL.md map to API stem USER / DATAMODEL and local ./<STEM>.md.

Bugfix: the first implementation compared the last four bytes to .md, but .md is three bytes, so names like USER.md never matched. Suffix handling now uses ".md".len().

Docs: clap help for the three subcommands and a short note in skills/hotdata/SKILL.md.

context list / --prefix are unchanged (prefix still matches server names as typed).

[sentry]

Codecov Report

❌ Patch coverage is 75.51020% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/context.rs	75.51%	12 Missing ⚠️

📢 Thoughts on this report? Let us know!

[claude]

Review

Blocking Issues

• src/context.rs lines 62-65: normalize_context_cli_name panics on user input that ends with a multi-byte UTF-8 character whose last bytes straddle the 3-byte .md boundary (e.g. "x𝕌" is 5 bytes → start = 2, which lands inside the 4-byte 𝕌 sequence → Rust panics with "byte index is not a char boundary").

Action Required

Replace the byte-level slice with a safe suffix check:

let lower = basename.to_ascii_lowercase();
if lower.ends_with(".md") {
    return basename[..basename.len() - MD_SUFFIX.len()].to_string();
}
basename.to_string()

lower.ends_with(".md") guarantees the last 3 bytes are the ASCII characters ., m, d. Because ASCII bytes are never UTF-8 continuation bytes, basename.len() - 3 is then guaranteed to be a valid character boundary, making the slice safe.

[claude]

Blocking issue from cycle 1 is fixed. The byte-level suffix check in normalize_context_cli_name is correct: it only slices basename[..i] after confirming the last three bytes are ASCII .md, which guarantees i is a valid UTF-8 char boundary. Tests cover the multi-byte edge cases directly.