feat(api): add random-trace-id flag to TraceFlags

[harshitt13]

Refs #104

What

Adds RANDOM_FLAG (0x02), isRandom and withRandom to TraceFlags, mirroring the existing sampled pair.

The flags byte already survived propagation via asByte, so an incoming 02 was re-emitted as 02 by accident of the whole byte being copied. What was missing was the ability to express the bit: nothing could ask whether an incoming trace ID is declared random, and the SDK had no supported way to set it on a root span context.

W3C Trace Context Level 2 makes preserving the bit across a continued trace a MUST. https://www.w3.org/TR/trace-context-2/#random-trace-id-flag

Notes

• withRandom does not verify randomness and cannot. Setting the flag is an assertion by the caller; enforcing it belongs to the SDK, which only sets it for a conforming ID generator. Documented on the method.
• Class doc now states that TraceFlags is a bit set and that flags == TraceFlags.sampled is wrong once a second bit exists.
• Changelog entry under Added, not Fixed. Nothing changed for existing callers, so this is new public surface under Keep a Changelog. Per the conventions comment the spec citation carries the audit provenance, and fix(api): Context-driven span creation and root span support (#83, #84) #118 set the precedent of an audit item landing its additive surface under Added.
• No TraceFlags.random constant. The class doc argues against having named whole-value constants to compare against.
• Out of scope: fromString leniency ([BUG] Trace API: TraceFlags.fromString coerces invalid input to 0x00 (API half) #92), and the mutable none and sampled statics, which I filed separately as #NNN.

Follow-up: whole-value comparisons in tests

Grepping == TraceFlags and equals(TraceFlags...) turns up 9 sites. Only 4 should change.

Wrong model, assert "is sampled" via whole-value equality: span_context_test.dart lines 73, 119, 168, 169.

Legitimate, leave alone, round-trip assertions where the whole byte is the subject: span_context_test.dart lines 56, 254, 261, 281, and span_context_create_test.dart:36.

The criterion is what the assertion is about. A round trip on withTraceFlags asserts the setter stored the byte you handed it. A sampling check asserts a predicate, and equality only answers it while no other bit is set.

The clearest case is line 150, a test named 'trace flags indicates sampling state' whose entire subject is the sampling predicate, tested by whole-value equality. Three of the four also carry a comment reading // Use accessor method instead of isSampled property directly above an == comparison, which has it backwards.

All 4 are inert today: they build their flags explicitly, so no random bit reaches them. They fail once root spans start emitting 03. git log -L puts them in the initial revision, so it is original habit rather than a later sweep, which means the SDK repo likely carries the same idiom from the same authorship. That is the urgent half, since that is where #132 makes root spans 03 for real.

Testing

Covers all four combinations of the two bits, both orders of application, that each with* preserves the other bit, and round trips through fromString/toString.