fix: resolve 7 latent issues and document 18 telemetry bugs

[deer]

Works through the latent issues backlog from docs/analysis.md, fixes 7 of the 10 items, and captures a new batch of telemetry bugs for a follow-up.

Fixes

• RetryFrequency — within() had inverted floor/ceiling guards (valid inputs threw); never() called maxRetriesOf(0) which also threw. Tests added for both.
• TextualRange.create() — checked start twice, never checked end. Test added.
• Table.getRow(int) — off-by-one: > rows.size() should be >= rows.size(). Test added.
• SchemaFactory — renamed getMashallingSchema → getMarshallingSchema and getUnmashallingSchemas → getUnmarshallingSchemas (missing 'r' in both).
• @Unmarshal — restricted @Target to CONSTRUCTOR only; register() now throws a clear error if @Unmarshal is found on a method instead of silently ignoring it.
• Progress.percentage() — changed return type from int to double (was truncating e.g. 33 instead of 33.3); also added division-by-zero guard for maximum == 0.

Docs

• docs/analysis.md — crossed off completed items, marked deferred Connection work, added status column to priority ranking.
• docs/telemetry-bugs.md (new) — 18 undocumented bugs across base-telemetry, base-telemetry-foundation, and base-telemetry-ansi, ranging from an always-NPE in TelemetryRecorder.diagnostic(Location, ...) to an off-by-one in the ANSI cursor movement that corrupts terminal output. Captured for a separate issue.

Not included

• Connection serialization / synchronized issues (intentionally deferred)
• Marshalling.GLOBAL_SCHEMA_FACTORY singleton (needs design discussion)
• Telemetry bug fixes (tracked in docs/telemetry-bugs.md)