fix: harden transport, JSON fidelity, and profile recovery

[francoischalifour]

The CLI had several correctness and reliability gaps at system boundaries:

• Streamed requests could hang while reading non-2xx response bodies.
• Peer-initiated stream aborts could be reported misleadingly as CLI timeouts.
• Typed API integers could lose precision or produce invalid JSON when they exceeded JavaScript’s safe range or contained leading zeroes.
• Human-readable query output could leak sensitive values from JSON string cells, while redaction could round unrelated large numeric identifiers.
• Configuration file I/O failures were sometimes treated as missing files.
• Profile rename/delete operations could leave credentials, directories, or active-profile state inconsistent after partial failures.
• Rollback failures were suppressed, making incomplete recovery difficult to diagnose.

Proposed solution

This branch hardens those boundaries by:

• Applying configured read deadlines while consuming both successful and error response streams.
• Distinguishing failures caused by the CLI’s timeout signal from peer/network aborts.
• Canonicalizing typed integer tokens before serializing them, while preserving their exact magnitude and digits in request bodies and query parameters.
• Preserving append JSON payloads after validation instead of parsing and reserializing them.
• Surfacing configuration filesystem failures as ConfigurationError while continuing to ignore only genuinely missing files.
• Replacing parse-and-reserialize query redaction with a source-preserving JSON walker that retains non-sensitive numeric lexemes, formatting, escaped keys, arrays, and nested values.
• Making profile secret moves transactional across source deletion failures.
• Restoring profile directories and configuration bytes after partial recursive deletion, including valid profiles with no config file.
• Reporting incomplete rollback with a ConfigurationError that retains the original failure as its cause and lists each failed recovery step.

Testing

Added regression coverage for:

• Large, negative, zero-padded, and query-string integer values.
• Stalled streamed error responses.
• Peer-aborted streams.
• Nested arrays and structured sensitive JSON values.
• Escaped sensitive keys and whitespace preservation.
• Malformed JSON redaction input.
• Configless profile restoration.
• Partial deletion and incomplete rollback diagnostics.

[albert20260301]

Reviewed the transport timeout boundaries, source-preserving JSON handling, and profile rollback paths. The implementation and regression coverage look sound; focused tests and type-aware lint pass locally.

[leonkenneth]

I think a clearer way would be to use a random id for config and secret prefixes, and just point to the id in the profile config.

Or just have profile ids be these random ids, and have a profile display-label (more or less the same)

[albert20260301]

The stable-ID follow-up passes focused profile/secrets/configure tests and type-aware lint locally. GitHub currently reports this branch as conflicting with main; please resolve that conflict so the required CI suite can run again before merge.